![\"Getting](\"https:\/\/www.kdnuggets.com\/wp-content\/uploads\/awan_getting_started_claude_agent_sdk_2.png\"\/)
![\"Getting](\"https:\/\/www.kdnuggets.com\/wp-content\/uploads\/awan_getting_started_claude_agent_sdk_2.png\")
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\n
#\u00a0<\/span>Introduction<\/h2>\n\u00a0
Uninterested in duct-taping scripts, instruments, and prompts collectively? The Claude Agent SDK<\/a><\/strong> enables you to flip your Claude Code<\/a><\/strong> \u201cplan \u2192 construct \u2192 run\u201d workflow into actual, programmable brokers, so you may automate duties, wire up instruments, and ship command line interface (CLI) apps with out tons of glue code. When you already like utilizing Claude<\/strong> within the terminal, this software program improvement equipment (SDK) provides you an identical vibe with correct construction, state, and extensibility.<\/p>\nOn this tutorial, you’ll arrange the Claude Agent SDK and construct a small, multi-tool CLI that chains steps end-to-end (plan \u2192 act \u2192 confirm). Alongside the best way, you will see learn how to register instruments, handle context, and orchestrate agent loops for native workflows like debugging, code technology, and deployment.
\u00a0<\/p>\n
#\u00a0<\/span>What’s the Claude Agent SDK?<\/h2>\n\u00a0
Anthropic<\/a><\/strong>‘s Claude Sonnet 4.5<\/strong> marks a major development in capabilities, that includes a state-of-the-art coding mannequin that excels in business benchmarks for reasoning, arithmetic, and long-context duties. This launch features a Chrome extension, a reminiscence instrument, and doc technology options. The standout part is the Claude Agent SDK<\/strong>, constructed on the muse of Claude Code<\/strong>.<\/p>\nThe Claude Agent SDK allows builders to create, prolong, and customise functions powered by Claude. It permits integration along with your native surroundings, granting Claude entry to your instruments and facilitating the orchestration of complicated workflows, together with coding, analysis, note-taking, and automation.<\/p>\n
\u00a0<\/p>\n
#\u00a0<\/span>Setting Up the Claude Agent SDK<\/h2>\n\u00a0
Earlier than constructing, be sure you’ve arrange each Claude Code CLI<\/strong> and the Claude Agent SDK<\/strong>.<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>1. Conditions<\/h4>\n\n- Python<\/a>:<\/strong> model 3.10 or larger.\n<\/li>\n
- Node.js<\/a>:<\/strong> model 18+ for the CLI.\n<\/li>\n
- Claude API Key<\/strong> or Anthropic account.\n<\/li>\n<\/ul>\n
\u00a0<\/p>\n
\/\/\u00a0<\/span>2. Set up Claude Code CLI<\/h4>\nWe are going to set up the Claude Code CLI on Home windows by typing the next command in PowerShell:<\/p>\n
\nirm https:\/\/claude.ai\/set up.ps1 | iex<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
Then add this path to your system surroundings:<\/p>\n
\u00a0<\/p>\n
Restart PowerShell and take a look at:<\/p>\n
\u00a0<\/p>\n
For different platforms, think about using the npm bundle supervisor:<\/p>\n
\nnpm i -g @anthropic-ai\/claude-code<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
After set up, kind claude<\/code> in your terminal to check in.<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Set up the Claude Agent SDK (Python)<\/h4>\nSet up the Claude Agent Python SDK utilizing the pip bundle supervisor.<\/p>\n
\npip set up claude-agent-sdk<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
When you get a CLINotFoundError<\/code>, make sure the Claude CLI is accurately put in and included in your PATH.<\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Constructing a Multi-Device App with the Claude Agent SDK<\/h2>\n\u00a0
On this part, we are going to construct the TrendSmith<\/strong> software, which tracks dwell market traits throughout varied industries, together with startups, AI, finance, and sustainability.<\/p>\nIt combines Claude Sonnet 4.5<\/strong>, WebSearch<\/strong>, WebFetch<\/strong>, and native storage<\/strong> instruments right into a single multi-agent system.<\/p>\nCreate the Python file trend_smith.py<\/code> and add the next code to it:<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>1. Imports & Fundamental Settings<\/h4>\nThis masses Python libraries, the Claude Agent SDK sorts, a tiny assist menu, the mannequin identify, and smooth grey textual content styling for standing traces.<\/p>\n
\nimport asyncio
\nimport os
\nimport re
\nimport sys
\nimport time
\nfrom datetime import datetime
\nfrom pathlib import Path
\n
\nfrom claude_agent_sdk import (
\n AssistantMessage,
\n ClaudeAgentOptions,
\n ClaudeSDKClient,
\n ResultMessage,
\n TextBlock,
\n ToolResultBlock,
\n ToolUseBlock,
\n)
\n
\nHELP = \"\"\"Instructions:
\n\/development Fast multi-source scan (auto-saves markdown)
\n\/scan Quick one-page scan
\n\/assist \/exit Assist \/ Stop
\n\"\"\"
\n
\nMODEL = os.getenv(\"CLAUDE_MODEL\", \"sonnet\") # e.g. \"sonnet-4.5\"
\nGRAY = \"\u000033[90m\"
\nRESET = \"\u000033[0m\"<\/query><\/topic><\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>2. System Prompt & Report Destination<\/h4>\nThis sets the \u201chouse rules\u201d for answers (fast, compact, consistent sections) and chooses a reports\/<\/code> folder next to your script for saved briefs.<\/p>\n\nSYS = \"\"\"You are TrendSmith, a fast, concise trend researcher.
\n- Finish quickly (~20 s).
\n- For \/trend: \u22641 WebSearch + \u22642 WebFetch from distinct domains.
\n- For \/scan: \u22641 WebFetch only.
\nReturn for \/trend:
\n TL;DR (1 line)
\n 3-5 Signals (short bullets)
\n Key Players, Risks, 30\/90-day Watchlist
\n Sources (markdown: **Title** -- URL)
\nReturn for \/scan: 5 bullets + TL;DR + Sources.
\nAfter finishing \/trend, the client will auto-save your full brief.
\n\"\"\"
\n
\nBASE = Path(__file__).parent
\nREPORTS = BASE \/ \"reports\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Saving Files Safely<\/h4>\nThese helpers make filenames safe, create folders if needed, and always try a home-folder fallback so your report still gets saved.<\/p>\n
\ndef _ts():
\n return datetime.now().strftime(\"%Y%m%d_%H%M\")
\n
\ndef _sanitize(s: str):
\n return re.sub(r\"[^w-.]+\", \"_\", s).strip(\"_\") or \"untitled\"
\n
\ndef _ensure_dir(p: Path):
\n strive:
\n p.mkdir(dad and mom=True, exist_ok=True)
\n besides Exception:
\n go
\n
\ndef _safe_write(path: Path, textual content: str) -> Path:
\n \"\"\"Write textual content to path; if listing\/permission fails, fall again to ~\/TrendSmith\/experiences.\"\"\"
\n strive:
\n _ensure_dir(path.mum or dad)
\n path.write_text(textual content, encoding=\"utf-8\")
\n return path
\n besides Exception:
\n home_reports = Path.dwelling() \/ \"TrendSmith\"https:\/\/www.kdnuggets.com\/\"experiences\"
\n _ensure_dir(home_reports)
\n fb = home_reports \/ path.identify
\n fb.write_text(textual content, encoding=\"utf-8\")
\n return fb
\n
\ndef save_report(subject: str, textual content: str) -> Path:
\n filename = f\"{_sanitize(subject)}_{_ts()}.md\"
\n goal = REPORTS \/ filename
\n return _safe_write(goal, textual content.strip() + \"n\")<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>4. Monitoring Every Run<\/h4>\nThis retains what you want for one request: streamed textual content, mannequin, instrument counts, token utilization, and timing, then resets cleanly earlier than the following request.<\/p>\n
\nclass State:
\n def __init__(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.last_cmd = None
\n self.last_topic = None
\n self.instruments = {}
\n self.t0 = 0.0
\n self.t1 = 0.0
\n
\n def reset(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.instruments = {}
\n self.t0 = time.perf_counter()
\n self.t1 = 0.0
\n
\ndef friendly_model(identify: str | None) -> str:
\n if not identify:
\n return MODEL
\n n = (identify or \"\").decrease()
\n if \"sonnet-4-5\" in n or \"sonnet_4_5\" in n:
\n return \"Claude 4.5 Sonnet\"
\n if \"sonnet\" in n:
\n return \"Claude Sonnet\"
\n if \"haiku\" in n:
\n return \"Claude Haiku\"
\n if \"opus\" in n:
\n return \"Claude Opus\"
\n return identify or \"Unknown\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>5. Quick Run Abstract<\/h4>\nThis prints a neat grey field to indicate the mannequin, tokens, instrument utilization, and period, with out mixing into your streamed content material.<\/p>\n
\ndef usage_footer(st: State, opts_model: str):
\n st.t1 = st.t1 or time.perf_counter()
\n dur = st.t1 - st.t0
\n utilization = st.utilization or {}
\n it = utilization.get(\"input_tokens\")
\n ot = utilization.get(\"output_tokens\")
\n complete = utilization.get(\"total_tokens\")
\n if complete is None and (it isn't None or ot is just not None):
\n complete = (it or 0) + (ot or 0)
\n tools_used = \", \".be part of(f\"{okay}\u00d7{v}\" for okay, v in st.instruments.objects()) or \"--\"
\n model_label = friendly_model(st.model_raw or opts_model)
\n
\n field = [
\n \"\u250c\u2500 Run Summary \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n f\"\u2502 Model: {model_label}\",
\n f\"\u2502 Tokens: {total if total is not None else '?'}\"
\n + (f\" (in={it if it is not None else '?'} | out={ot if ot is not None else '?'})\"
\n if (it is not None or ot is not None) else \"\"),
\n f\"\u2502 Tools: {tools_used}\",
\n f\"\u2502 Duration: {dur:.1f}s\",
\n \"\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n ]
\n print(GRAY + \"n\".be part of(field) + RESET, file=sys.stderr)<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>6. The Predominant Loop (All-in-One)<\/h4>\nThis begins the app, reads your command, asks the AI, streams the reply, saves \/development<\/code> experiences, and prints the abstract.<\/p>\n\nasync def most important():
\n \"\"\"Setup \u2192 REPL \u2192 parse \u2192 question\/stream \u2192 auto-save \u2192 abstract.\"\"\"
\n st = State()
\n _ensure_dir(REPORTS)
\n
\n opts = ClaudeAgentOptions(
\n mannequin=MODEL,
\n system_prompt=SYS,
\n allowed_tools=[\"WebFetch\", \"WebSearch\"],
\n )
\n
\n print(\"\ud83d\udcc8 TrendSmith nn\" + HELP)
\n
\n async with ClaudeSDKClient(choices=opts) as shopper:
\n whereas True:
\n # Learn enter
\n strive:
\n consumer = enter(\"nYou: \").strip()
\n besides (EOFError, KeyboardInterrupt):
\n print(\"nBye!\")
\n break
\n
\n if not consumer:
\n proceed
\n low = consumer.decrease()
\n
\n # Fundamental instructions
\n if low in {\"\/exit\", \"exit\", \"stop\"}:
\n print(\"Bye!\")
\n break
\n if low in {\"\/assist\", \"assist\"}:
\n print(HELP)
\n proceed
\n
\n # Parse right into a immediate
\n if low.startswith(\"\/development \"):
\n subject = consumer.cut up(\" \", 1)[1].strip().strip('\"')
\n if not subject:
\n print('e.g. \/development \"AI chip startups\"')
\n proceed
\n st.last_cmd, st.last_topic = \"development\", subject
\n immediate = f\"Run a quick development scan for '{subject}' following the output spec.\"
\n elif low.startswith(\"\/scan \"):
\n q = consumer.cut up(\" \", 1)[1].strip()
\n if not q:
\n print('e.g. \/scan \"AI {hardware} information\"')
\n proceed
\n st.last_cmd, st.last_topic = \"scan\", q
\n immediate = f\"Fast scan for '{q}' in underneath 10s (\u22641 WebFetch). Return 5 bullets + TL;DR + sources.\"
\n else:
\n st.last_cmd, st.last_topic = \"free\", None
\n immediate = consumer
\n
\n # Execute request and stream outcomes
\n st.reset()
\n print(f\"{GRAY}\u25b6 Working...{RESET}\")
\n strive:
\n await shopper.question(immediate)
\n besides Exception as e:
\n print(f\"{GRAY}\u274c Question error: {e}{RESET}\")
\n proceed
\n
\n strive:
\n async for m in shopper.receive_response():
\n if isinstance(m, AssistantMessage):
\n st.model_raw = st.model_raw or m.mannequin
\n for b in m.content material:
\n if isinstance(b, TextBlock):
\n st.temporary += b.textual content or \"\"
\n print(b.textual content or \"\", finish=\"\")
\n elif isinstance(b, ToolUseBlock):
\n identify = b.identify or \"Device\"
\n st.instruments[name] = st.instruments.get(identify, 0) + 1
\n print(f\"{GRAY}n\ud83d\udee0 Device: {identify}{RESET}\")
\n elif isinstance(b, ToolResultBlock):
\n go # quiet instrument payloads
\n elif isinstance(m, ResultMessage):
\n st.utilization = m.utilization or {}
\n st.value = m.total_cost_usd
\n besides Exception as e:
\n print(f\"{GRAY}n\u26a0 Stream error: {e}{RESET}\")
\n
\n # Auto-save development briefs and present the abstract
\n if st.last_cmd == \"development\" and st.temporary.strip():
\n strive:
\n saved_path = save_report(st.last_topic or \"development\", st.temporary)
\n print(f\"n{GRAY}\u2705 Auto-saved \u2192 {saved_path}{RESET}\")
\n besides Exception as e:
\n print(f\"{GRAY}\u26a0 Save error: {e}{RESET}\")
\n
\n st.t1 = time.perf_counter()
\n usage_footer(st, opts.mannequin)
\n
\nif __name__ == \"__main__\":
\n asyncio.run(most important())<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
#\u00a0<\/span>Testing the TrendSmith Utility<\/h2>\n\u00a0
We are going to now take a look at the app by operating the Python file. Here’s a fast recap on learn how to use the CLI software:<\/p>\n
\n- \/development “
“<\/topic><\/strong> \u2192 temporary multi-source scan, auto-saved to experiences\/_.md<\/ts><\/topic><\/code>.\n<\/li>\n- \/scan “
“<\/query><\/strong> \u2192 one-page fast scan (\u22641 WebFetch), prints solely.\n<\/li>\n- \/assist<\/strong> \u2192 exhibits instructions.\n<\/li>\n
- \/exit<\/strong> \u2192 quits.\n<\/li>\n<\/ul>\n
\u00a0<\/p>\n
![\"Getting](\"https:\/\/www.kdnuggets.com\/wp-content\/uploads\/awan_getting_started_claude_agent_sdk_5.png\"\/)
![\"Getting](\"https:\/\/www.kdnuggets.com\/wp-content\/uploads\/awan_getting_started_claude_agent_sdk_5.png\")
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe have now used the \/development<\/code> choice to seek for AI chip startups.<\/p>\n\n\/development \"AI chip startups\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
In consequence, the app has used varied search and net scraping instruments to collect info from completely different web sites.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nFinally, it has supplied the total response, auto-saved the report within the markdown file, and generated the utilization abstract. It value us $0.136.<\/p>\n
\u00a0<\/p>\n
![\"Getting](\"https:\/\/www.kdnuggets.com\/wp-content\/uploads\/awan_getting_started_claude_agent_sdk_6.png\"\/)
![\"Getting](\"https:\/\/www.kdnuggets.com\/wp-content\/uploads\/awan_getting_started_claude_agent_sdk_6.png\")
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nHere’s a preview of the saved Markdown report on the AI Chips Startups.<\/p>\n
\u00a0<\/p>\n
![\"Getting](\"https:\/\/www.kdnuggets.com\/wp-content\/uploads\/awan_getting_started_claude_agent_sdk_4.png\"\/)
![\"Getting](\"https:\/\/www.kdnuggets.com\/wp-content\/uploads\/awan_getting_started_claude_agent_sdk_4.png\")
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe are going to now take a look at the scanning choice and generate a abstract in regards to the subject utilizing an internet search.<\/p>\n
\u00a0<\/p>\n
It makes use of a easy net search and fetch instrument to generate a brief abstract on the subject.<\/p>\n
\u00a0<\/p>\n
![\"Getting](\"https:\/\/www.kdnuggets.com\/wp-content\/uploads\/awan_getting_started_claude_agent_sdk_1.png\"\/)
![\"Getting](\"https:\/\/www.kdnuggets.com\/wp-content\/uploads\/awan_getting_started_claude_agent_sdk_1.png\")
Picture by Writer<\/span><\/center><\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Last Ideas<\/h2>\n\u00a0
This app ran easily, and dealing with the Claude Agent SDK<\/strong> was genuinely enjoyable. If you’re already on the Claude Code<\/strong> plan, I extremely suggest attempting it to rework your day-to-day terminal workflow into dependable, repeatable agentic CLIs<\/strong>.<\/p>\nUse it to:<\/p>\n
\n- Automate widespread dev duties (debug, take a look at, deploy).\n<\/li>\n
- Script easy analytics or ops routines.\n<\/li>\n
- Bundle your circulation right into a reusable, shareable instrument.\n<\/li>\n<\/ul>\n
The SDK is an effective match for professionals who need stability<\/strong>, reproducibility<\/strong>, and low glue-code overhead<\/strong>. And sure, you may even ask Claude Code that will help you construct the agentic software itself with the SDK.
\u00a0
\u00a0<\/p>\nAbid Ali Awan<\/a><\/strong><\/b><\/a> (@1abidaliawan<\/a>) is a licensed knowledge scientist skilled who loves constructing machine studying fashions. Presently, he’s specializing in content material creation and writing technical blogs on machine studying and knowledge science applied sciences. Abid holds a Grasp’s diploma in expertise administration and a bachelor’s diploma in telecommunication engineering. His imaginative and prescient is to construct an AI product utilizing a graph neural community for college students combating psychological sickness.<\/p>\n<\/p><\/div>\n\n","protected":false},"excerpt":{"rendered":"Picture by Writer \u00a0 #\u00a0Introduction \u00a0Uninterested in duct-taping scripts, instruments, and prompts collectively? The Claude Agent SDK enables you to flip your Claude Code \u201cplan \u2192 construct \u2192 run\u201d workflow into actual, programmable brokers, so you may automate duties, wire up instruments, and ship command line interface (CLI) apps with out tons of glue code. […]<\/p>\n","protected":false},"author":2,"featured_media":9222,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[55],"tags":[75,458,721,2296],"class_list":["post-9220","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-machine-learning","tag-agent","tag-claude","tag-sdk","tag-started"],"_links":{"self":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9220"}],"version-history":[{"count":1,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions"}],"predecessor-version":[{"id":9221,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions\/9221"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/media\/9222"}],"wp:attachment":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9220"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9220"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9220"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
Uninterested in duct-taping scripts, instruments, and prompts collectively? The Claude Agent SDK<\/a><\/strong> enables you to flip your Claude Code<\/a><\/strong> \u201cplan \u2192 construct \u2192 run\u201d workflow into actual, programmable brokers, so you may automate duties, wire up instruments, and ship command line interface (CLI) apps with out tons of glue code. When you already like utilizing Claude<\/strong> within the terminal, this software program improvement equipment (SDK) provides you an identical vibe with correct construction, state, and extensibility.<\/p>\n
On this tutorial, you’ll arrange the Claude Agent SDK and construct a small, multi-tool CLI that chains steps end-to-end (plan \u2192 act \u2192 confirm). Alongside the best way, you will see learn how to register instruments, handle context, and orchestrate agent loops for native workflows like debugging, code technology, and deployment.
\u00a0<\/p>\n
#\u00a0<\/span>What’s the Claude Agent SDK?<\/h2>\n\u00a0
Anthropic<\/a><\/strong>‘s Claude Sonnet 4.5<\/strong> marks a major development in capabilities, that includes a state-of-the-art coding mannequin that excels in business benchmarks for reasoning, arithmetic, and long-context duties. This launch features a Chrome extension, a reminiscence instrument, and doc technology options. The standout part is the Claude Agent SDK<\/strong>, constructed on the muse of Claude Code<\/strong>.<\/p>\nThe Claude Agent SDK allows builders to create, prolong, and customise functions powered by Claude. It permits integration along with your native surroundings, granting Claude entry to your instruments and facilitating the orchestration of complicated workflows, together with coding, analysis, note-taking, and automation.<\/p>\n
\u00a0<\/p>\n
#\u00a0<\/span>Setting Up the Claude Agent SDK<\/h2>\n\u00a0
Earlier than constructing, be sure you’ve arrange each Claude Code CLI<\/strong> and the Claude Agent SDK<\/strong>.<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>1. Conditions<\/h4>\n\n- Python<\/a>:<\/strong> model 3.10 or larger.\n<\/li>\n
- Node.js<\/a>:<\/strong> model 18+ for the CLI.\n<\/li>\n
- Claude API Key<\/strong> or Anthropic account.\n<\/li>\n<\/ul>\n
\u00a0<\/p>\n
\/\/\u00a0<\/span>2. Set up Claude Code CLI<\/h4>\nWe are going to set up the Claude Code CLI on Home windows by typing the next command in PowerShell:<\/p>\n
\nirm https:\/\/claude.ai\/set up.ps1 | iex<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
Then add this path to your system surroundings:<\/p>\n
\u00a0<\/p>\n
Restart PowerShell and take a look at:<\/p>\n
\u00a0<\/p>\n
For different platforms, think about using the npm bundle supervisor:<\/p>\n
\nnpm i -g @anthropic-ai\/claude-code<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
After set up, kind claude<\/code> in your terminal to check in.<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Set up the Claude Agent SDK (Python)<\/h4>\nSet up the Claude Agent Python SDK utilizing the pip bundle supervisor.<\/p>\n
\npip set up claude-agent-sdk<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
When you get a CLINotFoundError<\/code>, make sure the Claude CLI is accurately put in and included in your PATH.<\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Constructing a Multi-Device App with the Claude Agent SDK<\/h2>\n\u00a0
On this part, we are going to construct the TrendSmith<\/strong> software, which tracks dwell market traits throughout varied industries, together with startups, AI, finance, and sustainability.<\/p>\nIt combines Claude Sonnet 4.5<\/strong>, WebSearch<\/strong>, WebFetch<\/strong>, and native storage<\/strong> instruments right into a single multi-agent system.<\/p>\nCreate the Python file trend_smith.py<\/code> and add the next code to it:<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>1. Imports & Fundamental Settings<\/h4>\nThis masses Python libraries, the Claude Agent SDK sorts, a tiny assist menu, the mannequin identify, and smooth grey textual content styling for standing traces.<\/p>\n
\nimport asyncio
\nimport os
\nimport re
\nimport sys
\nimport time
\nfrom datetime import datetime
\nfrom pathlib import Path
\n
\nfrom claude_agent_sdk import (
\n AssistantMessage,
\n ClaudeAgentOptions,
\n ClaudeSDKClient,
\n ResultMessage,
\n TextBlock,
\n ToolResultBlock,
\n ToolUseBlock,
\n)
\n
\nHELP = \"\"\"Instructions:
\n\/development Fast multi-source scan (auto-saves markdown)
\n\/scan Quick one-page scan
\n\/assist \/exit Assist \/ Stop
\n\"\"\"
\n
\nMODEL = os.getenv(\"CLAUDE_MODEL\", \"sonnet\") # e.g. \"sonnet-4.5\"
\nGRAY = \"\u000033[90m\"
\nRESET = \"\u000033[0m\"<\/query><\/topic><\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>2. System Prompt & Report Destination<\/h4>\nThis sets the \u201chouse rules\u201d for answers (fast, compact, consistent sections) and chooses a reports\/<\/code> folder next to your script for saved briefs.<\/p>\n\nSYS = \"\"\"You are TrendSmith, a fast, concise trend researcher.
\n- Finish quickly (~20 s).
\n- For \/trend: \u22641 WebSearch + \u22642 WebFetch from distinct domains.
\n- For \/scan: \u22641 WebFetch only.
\nReturn for \/trend:
\n TL;DR (1 line)
\n 3-5 Signals (short bullets)
\n Key Players, Risks, 30\/90-day Watchlist
\n Sources (markdown: **Title** -- URL)
\nReturn for \/scan: 5 bullets + TL;DR + Sources.
\nAfter finishing \/trend, the client will auto-save your full brief.
\n\"\"\"
\n
\nBASE = Path(__file__).parent
\nREPORTS = BASE \/ \"reports\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Saving Files Safely<\/h4>\nThese helpers make filenames safe, create folders if needed, and always try a home-folder fallback so your report still gets saved.<\/p>\n
\ndef _ts():
\n return datetime.now().strftime(\"%Y%m%d_%H%M\")
\n
\ndef _sanitize(s: str):
\n return re.sub(r\"[^w-.]+\", \"_\", s).strip(\"_\") or \"untitled\"
\n
\ndef _ensure_dir(p: Path):
\n strive:
\n p.mkdir(dad and mom=True, exist_ok=True)
\n besides Exception:
\n go
\n
\ndef _safe_write(path: Path, textual content: str) -> Path:
\n \"\"\"Write textual content to path; if listing\/permission fails, fall again to ~\/TrendSmith\/experiences.\"\"\"
\n strive:
\n _ensure_dir(path.mum or dad)
\n path.write_text(textual content, encoding=\"utf-8\")
\n return path
\n besides Exception:
\n home_reports = Path.dwelling() \/ \"TrendSmith\"https:\/\/www.kdnuggets.com\/\"experiences\"
\n _ensure_dir(home_reports)
\n fb = home_reports \/ path.identify
\n fb.write_text(textual content, encoding=\"utf-8\")
\n return fb
\n
\ndef save_report(subject: str, textual content: str) -> Path:
\n filename = f\"{_sanitize(subject)}_{_ts()}.md\"
\n goal = REPORTS \/ filename
\n return _safe_write(goal, textual content.strip() + \"n\")<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>4. Monitoring Every Run<\/h4>\nThis retains what you want for one request: streamed textual content, mannequin, instrument counts, token utilization, and timing, then resets cleanly earlier than the following request.<\/p>\n
\nclass State:
\n def __init__(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.last_cmd = None
\n self.last_topic = None
\n self.instruments = {}
\n self.t0 = 0.0
\n self.t1 = 0.0
\n
\n def reset(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.instruments = {}
\n self.t0 = time.perf_counter()
\n self.t1 = 0.0
\n
\ndef friendly_model(identify: str | None) -> str:
\n if not identify:
\n return MODEL
\n n = (identify or \"\").decrease()
\n if \"sonnet-4-5\" in n or \"sonnet_4_5\" in n:
\n return \"Claude 4.5 Sonnet\"
\n if \"sonnet\" in n:
\n return \"Claude Sonnet\"
\n if \"haiku\" in n:
\n return \"Claude Haiku\"
\n if \"opus\" in n:
\n return \"Claude Opus\"
\n return identify or \"Unknown\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>5. Quick Run Abstract<\/h4>\nThis prints a neat grey field to indicate the mannequin, tokens, instrument utilization, and period, with out mixing into your streamed content material.<\/p>\n
\ndef usage_footer(st: State, opts_model: str):
\n st.t1 = st.t1 or time.perf_counter()
\n dur = st.t1 - st.t0
\n utilization = st.utilization or {}
\n it = utilization.get(\"input_tokens\")
\n ot = utilization.get(\"output_tokens\")
\n complete = utilization.get(\"total_tokens\")
\n if complete is None and (it isn't None or ot is just not None):
\n complete = (it or 0) + (ot or 0)
\n tools_used = \", \".be part of(f\"{okay}\u00d7{v}\" for okay, v in st.instruments.objects()) or \"--\"
\n model_label = friendly_model(st.model_raw or opts_model)
\n
\n field = [
\n \"\u250c\u2500 Run Summary \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n f\"\u2502 Model: {model_label}\",
\n f\"\u2502 Tokens: {total if total is not None else '?'}\"
\n + (f\" (in={it if it is not None else '?'} | out={ot if ot is not None else '?'})\"
\n if (it is not None or ot is not None) else \"\"),
\n f\"\u2502 Tools: {tools_used}\",
\n f\"\u2502 Duration: {dur:.1f}s\",
\n \"\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n ]
\n print(GRAY + \"n\".be part of(field) + RESET, file=sys.stderr)<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>6. The Predominant Loop (All-in-One)<\/h4>\nThis begins the app, reads your command, asks the AI, streams the reply, saves \/development<\/code> experiences, and prints the abstract.<\/p>\n\nasync def most important():
\n \"\"\"Setup \u2192 REPL \u2192 parse \u2192 question\/stream \u2192 auto-save \u2192 abstract.\"\"\"
\n st = State()
\n _ensure_dir(REPORTS)
\n
\n opts = ClaudeAgentOptions(
\n mannequin=MODEL,
\n system_prompt=SYS,
\n allowed_tools=[\"WebFetch\", \"WebSearch\"],
\n )
\n
\n print(\"\ud83d\udcc8 TrendSmith nn\" + HELP)
\n
\n async with ClaudeSDKClient(choices=opts) as shopper:
\n whereas True:
\n # Learn enter
\n strive:
\n consumer = enter(\"nYou: \").strip()
\n besides (EOFError, KeyboardInterrupt):
\n print(\"nBye!\")
\n break
\n
\n if not consumer:
\n proceed
\n low = consumer.decrease()
\n
\n # Fundamental instructions
\n if low in {\"\/exit\", \"exit\", \"stop\"}:
\n print(\"Bye!\")
\n break
\n if low in {\"\/assist\", \"assist\"}:
\n print(HELP)
\n proceed
\n
\n # Parse right into a immediate
\n if low.startswith(\"\/development \"):
\n subject = consumer.cut up(\" \", 1)[1].strip().strip('\"')
\n if not subject:
\n print('e.g. \/development \"AI chip startups\"')
\n proceed
\n st.last_cmd, st.last_topic = \"development\", subject
\n immediate = f\"Run a quick development scan for '{subject}' following the output spec.\"
\n elif low.startswith(\"\/scan \"):
\n q = consumer.cut up(\" \", 1)[1].strip()
\n if not q:
\n print('e.g. \/scan \"AI {hardware} information\"')
\n proceed
\n st.last_cmd, st.last_topic = \"scan\", q
\n immediate = f\"Fast scan for '{q}' in underneath 10s (\u22641 WebFetch). Return 5 bullets + TL;DR + sources.\"
\n else:
\n st.last_cmd, st.last_topic = \"free\", None
\n immediate = consumer
\n
\n # Execute request and stream outcomes
\n st.reset()
\n print(f\"{GRAY}\u25b6 Working...{RESET}\")
\n strive:
\n await shopper.question(immediate)
\n besides Exception as e:
\n print(f\"{GRAY}\u274c Question error: {e}{RESET}\")
\n proceed
\n
\n strive:
\n async for m in shopper.receive_response():
\n if isinstance(m, AssistantMessage):
\n st.model_raw = st.model_raw or m.mannequin
\n for b in m.content material:
\n if isinstance(b, TextBlock):
\n st.temporary += b.textual content or \"\"
\n print(b.textual content or \"\", finish=\"\")
\n elif isinstance(b, ToolUseBlock):
\n identify = b.identify or \"Device\"
\n st.instruments[name] = st.instruments.get(identify, 0) + 1
\n print(f\"{GRAY}n\ud83d\udee0 Device: {identify}{RESET}\")
\n elif isinstance(b, ToolResultBlock):
\n go # quiet instrument payloads
\n elif isinstance(m, ResultMessage):
\n st.utilization = m.utilization or {}
\n st.value = m.total_cost_usd
\n besides Exception as e:
\n print(f\"{GRAY}n\u26a0 Stream error: {e}{RESET}\")
\n
\n # Auto-save development briefs and present the abstract
\n if st.last_cmd == \"development\" and st.temporary.strip():
\n strive:
\n saved_path = save_report(st.last_topic or \"development\", st.temporary)
\n print(f\"n{GRAY}\u2705 Auto-saved \u2192 {saved_path}{RESET}\")
\n besides Exception as e:
\n print(f\"{GRAY}\u26a0 Save error: {e}{RESET}\")
\n
\n st.t1 = time.perf_counter()
\n usage_footer(st, opts.mannequin)
\n
\nif __name__ == \"__main__\":
\n asyncio.run(most important())<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
#\u00a0<\/span>Testing the TrendSmith Utility<\/h2>\n\u00a0
We are going to now take a look at the app by operating the Python file. Here’s a fast recap on learn how to use the CLI software:<\/p>\n
\n- \/development “
“<\/topic><\/strong> \u2192 temporary multi-source scan, auto-saved to experiences\/_.md<\/ts><\/topic><\/code>.\n<\/li>\n- \/scan “
“<\/query><\/strong> \u2192 one-page fast scan (\u22641 WebFetch), prints solely.\n<\/li>\n- \/assist<\/strong> \u2192 exhibits instructions.\n<\/li>\n
- \/exit<\/strong> \u2192 quits.\n<\/li>\n<\/ul>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe have now used the \/development<\/code> choice to seek for AI chip startups.<\/p>\n\n\/development \"AI chip startups\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
In consequence, the app has used varied search and net scraping instruments to collect info from completely different web sites.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nFinally, it has supplied the total response, auto-saved the report within the markdown file, and generated the utilization abstract. It value us $0.136.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nHere’s a preview of the saved Markdown report on the AI Chips Startups.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe are going to now take a look at the scanning choice and generate a abstract in regards to the subject utilizing an internet search.<\/p>\n
\u00a0<\/p>\n
It makes use of a easy net search and fetch instrument to generate a brief abstract on the subject.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center><\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Last Ideas<\/h2>\n\u00a0
This app ran easily, and dealing with the Claude Agent SDK<\/strong> was genuinely enjoyable. If you’re already on the Claude Code<\/strong> plan, I extremely suggest attempting it to rework your day-to-day terminal workflow into dependable, repeatable agentic CLIs<\/strong>.<\/p>\nUse it to:<\/p>\n
\n- Automate widespread dev duties (debug, take a look at, deploy).\n<\/li>\n
- Script easy analytics or ops routines.\n<\/li>\n
- Bundle your circulation right into a reusable, shareable instrument.\n<\/li>\n<\/ul>\n
The SDK is an effective match for professionals who need stability<\/strong>, reproducibility<\/strong>, and low glue-code overhead<\/strong>. And sure, you may even ask Claude Code that will help you construct the agentic software itself with the SDK.
\u00a0
\u00a0<\/p>\nAbid Ali Awan<\/a><\/strong><\/b><\/a> (@1abidaliawan<\/a>) is a licensed knowledge scientist skilled who loves constructing machine studying fashions. Presently, he’s specializing in content material creation and writing technical blogs on machine studying and knowledge science applied sciences. Abid holds a Grasp’s diploma in expertise administration and a bachelor’s diploma in telecommunication engineering. His imaginative and prescient is to construct an AI product utilizing a graph neural community for college students combating psychological sickness.<\/p>\n<\/p><\/div>\n\n","protected":false},"excerpt":{"rendered":"Picture by Writer \u00a0 #\u00a0Introduction \u00a0Uninterested in duct-taping scripts, instruments, and prompts collectively? The Claude Agent SDK enables you to flip your Claude Code \u201cplan \u2192 construct \u2192 run\u201d workflow into actual, programmable brokers, so you may automate duties, wire up instruments, and ship command line interface (CLI) apps with out tons of glue code. […]<\/p>\n","protected":false},"author":2,"featured_media":9222,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[55],"tags":[75,458,721,2296],"class_list":["post-9220","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-machine-learning","tag-agent","tag-claude","tag-sdk","tag-started"],"_links":{"self":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9220"}],"version-history":[{"count":1,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions"}],"predecessor-version":[{"id":9221,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions\/9221"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/media\/9222"}],"wp:attachment":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9220"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9220"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9220"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
Anthropic<\/a><\/strong>‘s Claude Sonnet 4.5<\/strong> marks a major development in capabilities, that includes a state-of-the-art coding mannequin that excels in business benchmarks for reasoning, arithmetic, and long-context duties. This launch features a Chrome extension, a reminiscence instrument, and doc technology options. The standout part is the Claude Agent SDK<\/strong>, constructed on the muse of Claude Code<\/strong>.<\/p>\n
The Claude Agent SDK allows builders to create, prolong, and customise functions powered by Claude. It permits integration along with your native surroundings, granting Claude entry to your instruments and facilitating the orchestration of complicated workflows, together with coding, analysis, note-taking, and automation.<\/p>\n
\u00a0<\/p>\n
#\u00a0<\/span>Setting Up the Claude Agent SDK<\/h2>\n\u00a0
Earlier than constructing, be sure you’ve arrange each Claude Code CLI<\/strong> and the Claude Agent SDK<\/strong>.<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>1. Conditions<\/h4>\n\n- Python<\/a>:<\/strong> model 3.10 or larger.\n<\/li>\n
- Node.js<\/a>:<\/strong> model 18+ for the CLI.\n<\/li>\n
- Claude API Key<\/strong> or Anthropic account.\n<\/li>\n<\/ul>\n
\u00a0<\/p>\n
\/\/\u00a0<\/span>2. Set up Claude Code CLI<\/h4>\nWe are going to set up the Claude Code CLI on Home windows by typing the next command in PowerShell:<\/p>\n
\nirm https:\/\/claude.ai\/set up.ps1 | iex<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
Then add this path to your system surroundings:<\/p>\n
\u00a0<\/p>\n
Restart PowerShell and take a look at:<\/p>\n
\u00a0<\/p>\n
For different platforms, think about using the npm bundle supervisor:<\/p>\n
\nnpm i -g @anthropic-ai\/claude-code<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
After set up, kind claude<\/code> in your terminal to check in.<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Set up the Claude Agent SDK (Python)<\/h4>\nSet up the Claude Agent Python SDK utilizing the pip bundle supervisor.<\/p>\n
\npip set up claude-agent-sdk<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
When you get a CLINotFoundError<\/code>, make sure the Claude CLI is accurately put in and included in your PATH.<\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Constructing a Multi-Device App with the Claude Agent SDK<\/h2>\n\u00a0
On this part, we are going to construct the TrendSmith<\/strong> software, which tracks dwell market traits throughout varied industries, together with startups, AI, finance, and sustainability.<\/p>\nIt combines Claude Sonnet 4.5<\/strong>, WebSearch<\/strong>, WebFetch<\/strong>, and native storage<\/strong> instruments right into a single multi-agent system.<\/p>\nCreate the Python file trend_smith.py<\/code> and add the next code to it:<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>1. Imports & Fundamental Settings<\/h4>\nThis masses Python libraries, the Claude Agent SDK sorts, a tiny assist menu, the mannequin identify, and smooth grey textual content styling for standing traces.<\/p>\n
\nimport asyncio
\nimport os
\nimport re
\nimport sys
\nimport time
\nfrom datetime import datetime
\nfrom pathlib import Path
\n
\nfrom claude_agent_sdk import (
\n AssistantMessage,
\n ClaudeAgentOptions,
\n ClaudeSDKClient,
\n ResultMessage,
\n TextBlock,
\n ToolResultBlock,
\n ToolUseBlock,
\n)
\n
\nHELP = \"\"\"Instructions:
\n\/development Fast multi-source scan (auto-saves markdown)
\n\/scan Quick one-page scan
\n\/assist \/exit Assist \/ Stop
\n\"\"\"
\n
\nMODEL = os.getenv(\"CLAUDE_MODEL\", \"sonnet\") # e.g. \"sonnet-4.5\"
\nGRAY = \"\u000033[90m\"
\nRESET = \"\u000033[0m\"<\/query><\/topic><\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>2. System Prompt & Report Destination<\/h4>\nThis sets the \u201chouse rules\u201d for answers (fast, compact, consistent sections) and chooses a reports\/<\/code> folder next to your script for saved briefs.<\/p>\n\nSYS = \"\"\"You are TrendSmith, a fast, concise trend researcher.
\n- Finish quickly (~20 s).
\n- For \/trend: \u22641 WebSearch + \u22642 WebFetch from distinct domains.
\n- For \/scan: \u22641 WebFetch only.
\nReturn for \/trend:
\n TL;DR (1 line)
\n 3-5 Signals (short bullets)
\n Key Players, Risks, 30\/90-day Watchlist
\n Sources (markdown: **Title** -- URL)
\nReturn for \/scan: 5 bullets + TL;DR + Sources.
\nAfter finishing \/trend, the client will auto-save your full brief.
\n\"\"\"
\n
\nBASE = Path(__file__).parent
\nREPORTS = BASE \/ \"reports\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Saving Files Safely<\/h4>\nThese helpers make filenames safe, create folders if needed, and always try a home-folder fallback so your report still gets saved.<\/p>\n
\ndef _ts():
\n return datetime.now().strftime(\"%Y%m%d_%H%M\")
\n
\ndef _sanitize(s: str):
\n return re.sub(r\"[^w-.]+\", \"_\", s).strip(\"_\") or \"untitled\"
\n
\ndef _ensure_dir(p: Path):
\n strive:
\n p.mkdir(dad and mom=True, exist_ok=True)
\n besides Exception:
\n go
\n
\ndef _safe_write(path: Path, textual content: str) -> Path:
\n \"\"\"Write textual content to path; if listing\/permission fails, fall again to ~\/TrendSmith\/experiences.\"\"\"
\n strive:
\n _ensure_dir(path.mum or dad)
\n path.write_text(textual content, encoding=\"utf-8\")
\n return path
\n besides Exception:
\n home_reports = Path.dwelling() \/ \"TrendSmith\"https:\/\/www.kdnuggets.com\/\"experiences\"
\n _ensure_dir(home_reports)
\n fb = home_reports \/ path.identify
\n fb.write_text(textual content, encoding=\"utf-8\")
\n return fb
\n
\ndef save_report(subject: str, textual content: str) -> Path:
\n filename = f\"{_sanitize(subject)}_{_ts()}.md\"
\n goal = REPORTS \/ filename
\n return _safe_write(goal, textual content.strip() + \"n\")<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>4. Monitoring Every Run<\/h4>\nThis retains what you want for one request: streamed textual content, mannequin, instrument counts, token utilization, and timing, then resets cleanly earlier than the following request.<\/p>\n
\nclass State:
\n def __init__(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.last_cmd = None
\n self.last_topic = None
\n self.instruments = {}
\n self.t0 = 0.0
\n self.t1 = 0.0
\n
\n def reset(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.instruments = {}
\n self.t0 = time.perf_counter()
\n self.t1 = 0.0
\n
\ndef friendly_model(identify: str | None) -> str:
\n if not identify:
\n return MODEL
\n n = (identify or \"\").decrease()
\n if \"sonnet-4-5\" in n or \"sonnet_4_5\" in n:
\n return \"Claude 4.5 Sonnet\"
\n if \"sonnet\" in n:
\n return \"Claude Sonnet\"
\n if \"haiku\" in n:
\n return \"Claude Haiku\"
\n if \"opus\" in n:
\n return \"Claude Opus\"
\n return identify or \"Unknown\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>5. Quick Run Abstract<\/h4>\nThis prints a neat grey field to indicate the mannequin, tokens, instrument utilization, and period, with out mixing into your streamed content material.<\/p>\n
\ndef usage_footer(st: State, opts_model: str):
\n st.t1 = st.t1 or time.perf_counter()
\n dur = st.t1 - st.t0
\n utilization = st.utilization or {}
\n it = utilization.get(\"input_tokens\")
\n ot = utilization.get(\"output_tokens\")
\n complete = utilization.get(\"total_tokens\")
\n if complete is None and (it isn't None or ot is just not None):
\n complete = (it or 0) + (ot or 0)
\n tools_used = \", \".be part of(f\"{okay}\u00d7{v}\" for okay, v in st.instruments.objects()) or \"--\"
\n model_label = friendly_model(st.model_raw or opts_model)
\n
\n field = [
\n \"\u250c\u2500 Run Summary \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n f\"\u2502 Model: {model_label}\",
\n f\"\u2502 Tokens: {total if total is not None else '?'}\"
\n + (f\" (in={it if it is not None else '?'} | out={ot if ot is not None else '?'})\"
\n if (it is not None or ot is not None) else \"\"),
\n f\"\u2502 Tools: {tools_used}\",
\n f\"\u2502 Duration: {dur:.1f}s\",
\n \"\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n ]
\n print(GRAY + \"n\".be part of(field) + RESET, file=sys.stderr)<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>6. The Predominant Loop (All-in-One)<\/h4>\nThis begins the app, reads your command, asks the AI, streams the reply, saves \/development<\/code> experiences, and prints the abstract.<\/p>\n\nasync def most important():
\n \"\"\"Setup \u2192 REPL \u2192 parse \u2192 question\/stream \u2192 auto-save \u2192 abstract.\"\"\"
\n st = State()
\n _ensure_dir(REPORTS)
\n
\n opts = ClaudeAgentOptions(
\n mannequin=MODEL,
\n system_prompt=SYS,
\n allowed_tools=[\"WebFetch\", \"WebSearch\"],
\n )
\n
\n print(\"\ud83d\udcc8 TrendSmith nn\" + HELP)
\n
\n async with ClaudeSDKClient(choices=opts) as shopper:
\n whereas True:
\n # Learn enter
\n strive:
\n consumer = enter(\"nYou: \").strip()
\n besides (EOFError, KeyboardInterrupt):
\n print(\"nBye!\")
\n break
\n
\n if not consumer:
\n proceed
\n low = consumer.decrease()
\n
\n # Fundamental instructions
\n if low in {\"\/exit\", \"exit\", \"stop\"}:
\n print(\"Bye!\")
\n break
\n if low in {\"\/assist\", \"assist\"}:
\n print(HELP)
\n proceed
\n
\n # Parse right into a immediate
\n if low.startswith(\"\/development \"):
\n subject = consumer.cut up(\" \", 1)[1].strip().strip('\"')
\n if not subject:
\n print('e.g. \/development \"AI chip startups\"')
\n proceed
\n st.last_cmd, st.last_topic = \"development\", subject
\n immediate = f\"Run a quick development scan for '{subject}' following the output spec.\"
\n elif low.startswith(\"\/scan \"):
\n q = consumer.cut up(\" \", 1)[1].strip()
\n if not q:
\n print('e.g. \/scan \"AI {hardware} information\"')
\n proceed
\n st.last_cmd, st.last_topic = \"scan\", q
\n immediate = f\"Fast scan for '{q}' in underneath 10s (\u22641 WebFetch). Return 5 bullets + TL;DR + sources.\"
\n else:
\n st.last_cmd, st.last_topic = \"free\", None
\n immediate = consumer
\n
\n # Execute request and stream outcomes
\n st.reset()
\n print(f\"{GRAY}\u25b6 Working...{RESET}\")
\n strive:
\n await shopper.question(immediate)
\n besides Exception as e:
\n print(f\"{GRAY}\u274c Question error: {e}{RESET}\")
\n proceed
\n
\n strive:
\n async for m in shopper.receive_response():
\n if isinstance(m, AssistantMessage):
\n st.model_raw = st.model_raw or m.mannequin
\n for b in m.content material:
\n if isinstance(b, TextBlock):
\n st.temporary += b.textual content or \"\"
\n print(b.textual content or \"\", finish=\"\")
\n elif isinstance(b, ToolUseBlock):
\n identify = b.identify or \"Device\"
\n st.instruments[name] = st.instruments.get(identify, 0) + 1
\n print(f\"{GRAY}n\ud83d\udee0 Device: {identify}{RESET}\")
\n elif isinstance(b, ToolResultBlock):
\n go # quiet instrument payloads
\n elif isinstance(m, ResultMessage):
\n st.utilization = m.utilization or {}
\n st.value = m.total_cost_usd
\n besides Exception as e:
\n print(f\"{GRAY}n\u26a0 Stream error: {e}{RESET}\")
\n
\n # Auto-save development briefs and present the abstract
\n if st.last_cmd == \"development\" and st.temporary.strip():
\n strive:
\n saved_path = save_report(st.last_topic or \"development\", st.temporary)
\n print(f\"n{GRAY}\u2705 Auto-saved \u2192 {saved_path}{RESET}\")
\n besides Exception as e:
\n print(f\"{GRAY}\u26a0 Save error: {e}{RESET}\")
\n
\n st.t1 = time.perf_counter()
\n usage_footer(st, opts.mannequin)
\n
\nif __name__ == \"__main__\":
\n asyncio.run(most important())<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
#\u00a0<\/span>Testing the TrendSmith Utility<\/h2>\n\u00a0
We are going to now take a look at the app by operating the Python file. Here’s a fast recap on learn how to use the CLI software:<\/p>\n
\n- \/development “
“<\/topic><\/strong> \u2192 temporary multi-source scan, auto-saved to experiences\/_.md<\/ts><\/topic><\/code>.\n<\/li>\n- \/scan “
“<\/query><\/strong> \u2192 one-page fast scan (\u22641 WebFetch), prints solely.\n<\/li>\n- \/assist<\/strong> \u2192 exhibits instructions.\n<\/li>\n
- \/exit<\/strong> \u2192 quits.\n<\/li>\n<\/ul>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe have now used the \/development<\/code> choice to seek for AI chip startups.<\/p>\n\n\/development \"AI chip startups\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
In consequence, the app has used varied search and net scraping instruments to collect info from completely different web sites.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nFinally, it has supplied the total response, auto-saved the report within the markdown file, and generated the utilization abstract. It value us $0.136.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nHere’s a preview of the saved Markdown report on the AI Chips Startups.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe are going to now take a look at the scanning choice and generate a abstract in regards to the subject utilizing an internet search.<\/p>\n
\u00a0<\/p>\n
It makes use of a easy net search and fetch instrument to generate a brief abstract on the subject.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center><\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Last Ideas<\/h2>\n\u00a0
This app ran easily, and dealing with the Claude Agent SDK<\/strong> was genuinely enjoyable. If you’re already on the Claude Code<\/strong> plan, I extremely suggest attempting it to rework your day-to-day terminal workflow into dependable, repeatable agentic CLIs<\/strong>.<\/p>\nUse it to:<\/p>\n
\n- Automate widespread dev duties (debug, take a look at, deploy).\n<\/li>\n
- Script easy analytics or ops routines.\n<\/li>\n
- Bundle your circulation right into a reusable, shareable instrument.\n<\/li>\n<\/ul>\n
The SDK is an effective match for professionals who need stability<\/strong>, reproducibility<\/strong>, and low glue-code overhead<\/strong>. And sure, you may even ask Claude Code that will help you construct the agentic software itself with the SDK.
\u00a0
\u00a0<\/p>\nAbid Ali Awan<\/a><\/strong><\/b><\/a> (@1abidaliawan<\/a>) is a licensed knowledge scientist skilled who loves constructing machine studying fashions. Presently, he’s specializing in content material creation and writing technical blogs on machine studying and knowledge science applied sciences. Abid holds a Grasp’s diploma in expertise administration and a bachelor’s diploma in telecommunication engineering. His imaginative and prescient is to construct an AI product utilizing a graph neural community for college students combating psychological sickness.<\/p>\n<\/p><\/div>\n\n","protected":false},"excerpt":{"rendered":"Picture by Writer \u00a0 #\u00a0Introduction \u00a0Uninterested in duct-taping scripts, instruments, and prompts collectively? The Claude Agent SDK enables you to flip your Claude Code \u201cplan \u2192 construct \u2192 run\u201d workflow into actual, programmable brokers, so you may automate duties, wire up instruments, and ship command line interface (CLI) apps with out tons of glue code. […]<\/p>\n","protected":false},"author":2,"featured_media":9222,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[55],"tags":[75,458,721,2296],"class_list":["post-9220","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-machine-learning","tag-agent","tag-claude","tag-sdk","tag-started"],"_links":{"self":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9220"}],"version-history":[{"count":1,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions"}],"predecessor-version":[{"id":9221,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions\/9221"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/media\/9222"}],"wp:attachment":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9220"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9220"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9220"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
Earlier than constructing, be sure you’ve arrange each Claude Code CLI<\/strong> and the Claude Agent SDK<\/strong>.<\/p>\n
\u00a0<\/p>\n
\/\/\u00a0<\/span>1. Conditions<\/h4>\n\n- Python<\/a>:<\/strong> model 3.10 or larger.\n<\/li>\n
- Node.js<\/a>:<\/strong> model 18+ for the CLI.\n<\/li>\n
- Claude API Key<\/strong> or Anthropic account.\n<\/li>\n<\/ul>\n
\u00a0<\/p>\n
\/\/\u00a0<\/span>2. Set up Claude Code CLI<\/h4>\nWe are going to set up the Claude Code CLI on Home windows by typing the next command in PowerShell:<\/p>\n
\nirm https:\/\/claude.ai\/set up.ps1 | iex<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
Then add this path to your system surroundings:<\/p>\n
\u00a0<\/p>\n
Restart PowerShell and take a look at:<\/p>\n
\u00a0<\/p>\n
For different platforms, think about using the npm bundle supervisor:<\/p>\n
\nnpm i -g @anthropic-ai\/claude-code<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
After set up, kind claude<\/code> in your terminal to check in.<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Set up the Claude Agent SDK (Python)<\/h4>\nSet up the Claude Agent Python SDK utilizing the pip bundle supervisor.<\/p>\n
\npip set up claude-agent-sdk<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
When you get a CLINotFoundError<\/code>, make sure the Claude CLI is accurately put in and included in your PATH.<\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Constructing a Multi-Device App with the Claude Agent SDK<\/h2>\n\u00a0
On this part, we are going to construct the TrendSmith<\/strong> software, which tracks dwell market traits throughout varied industries, together with startups, AI, finance, and sustainability.<\/p>\nIt combines Claude Sonnet 4.5<\/strong>, WebSearch<\/strong>, WebFetch<\/strong>, and native storage<\/strong> instruments right into a single multi-agent system.<\/p>\nCreate the Python file trend_smith.py<\/code> and add the next code to it:<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>1. Imports & Fundamental Settings<\/h4>\nThis masses Python libraries, the Claude Agent SDK sorts, a tiny assist menu, the mannequin identify, and smooth grey textual content styling for standing traces.<\/p>\n
\nimport asyncio
\nimport os
\nimport re
\nimport sys
\nimport time
\nfrom datetime import datetime
\nfrom pathlib import Path
\n
\nfrom claude_agent_sdk import (
\n AssistantMessage,
\n ClaudeAgentOptions,
\n ClaudeSDKClient,
\n ResultMessage,
\n TextBlock,
\n ToolResultBlock,
\n ToolUseBlock,
\n)
\n
\nHELP = \"\"\"Instructions:
\n\/development Fast multi-source scan (auto-saves markdown)
\n\/scan Quick one-page scan
\n\/assist \/exit Assist \/ Stop
\n\"\"\"
\n
\nMODEL = os.getenv(\"CLAUDE_MODEL\", \"sonnet\") # e.g. \"sonnet-4.5\"
\nGRAY = \"\u000033[90m\"
\nRESET = \"\u000033[0m\"<\/query><\/topic><\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>2. System Prompt & Report Destination<\/h4>\nThis sets the \u201chouse rules\u201d for answers (fast, compact, consistent sections) and chooses a reports\/<\/code> folder next to your script for saved briefs.<\/p>\n\nSYS = \"\"\"You are TrendSmith, a fast, concise trend researcher.
\n- Finish quickly (~20 s).
\n- For \/trend: \u22641 WebSearch + \u22642 WebFetch from distinct domains.
\n- For \/scan: \u22641 WebFetch only.
\nReturn for \/trend:
\n TL;DR (1 line)
\n 3-5 Signals (short bullets)
\n Key Players, Risks, 30\/90-day Watchlist
\n Sources (markdown: **Title** -- URL)
\nReturn for \/scan: 5 bullets + TL;DR + Sources.
\nAfter finishing \/trend, the client will auto-save your full brief.
\n\"\"\"
\n
\nBASE = Path(__file__).parent
\nREPORTS = BASE \/ \"reports\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Saving Files Safely<\/h4>\nThese helpers make filenames safe, create folders if needed, and always try a home-folder fallback so your report still gets saved.<\/p>\n
\ndef _ts():
\n return datetime.now().strftime(\"%Y%m%d_%H%M\")
\n
\ndef _sanitize(s: str):
\n return re.sub(r\"[^w-.]+\", \"_\", s).strip(\"_\") or \"untitled\"
\n
\ndef _ensure_dir(p: Path):
\n strive:
\n p.mkdir(dad and mom=True, exist_ok=True)
\n besides Exception:
\n go
\n
\ndef _safe_write(path: Path, textual content: str) -> Path:
\n \"\"\"Write textual content to path; if listing\/permission fails, fall again to ~\/TrendSmith\/experiences.\"\"\"
\n strive:
\n _ensure_dir(path.mum or dad)
\n path.write_text(textual content, encoding=\"utf-8\")
\n return path
\n besides Exception:
\n home_reports = Path.dwelling() \/ \"TrendSmith\"https:\/\/www.kdnuggets.com\/\"experiences\"
\n _ensure_dir(home_reports)
\n fb = home_reports \/ path.identify
\n fb.write_text(textual content, encoding=\"utf-8\")
\n return fb
\n
\ndef save_report(subject: str, textual content: str) -> Path:
\n filename = f\"{_sanitize(subject)}_{_ts()}.md\"
\n goal = REPORTS \/ filename
\n return _safe_write(goal, textual content.strip() + \"n\")<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>4. Monitoring Every Run<\/h4>\nThis retains what you want for one request: streamed textual content, mannequin, instrument counts, token utilization, and timing, then resets cleanly earlier than the following request.<\/p>\n
\nclass State:
\n def __init__(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.last_cmd = None
\n self.last_topic = None
\n self.instruments = {}
\n self.t0 = 0.0
\n self.t1 = 0.0
\n
\n def reset(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.instruments = {}
\n self.t0 = time.perf_counter()
\n self.t1 = 0.0
\n
\ndef friendly_model(identify: str | None) -> str:
\n if not identify:
\n return MODEL
\n n = (identify or \"\").decrease()
\n if \"sonnet-4-5\" in n or \"sonnet_4_5\" in n:
\n return \"Claude 4.5 Sonnet\"
\n if \"sonnet\" in n:
\n return \"Claude Sonnet\"
\n if \"haiku\" in n:
\n return \"Claude Haiku\"
\n if \"opus\" in n:
\n return \"Claude Opus\"
\n return identify or \"Unknown\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>5. Quick Run Abstract<\/h4>\nThis prints a neat grey field to indicate the mannequin, tokens, instrument utilization, and period, with out mixing into your streamed content material.<\/p>\n
\ndef usage_footer(st: State, opts_model: str):
\n st.t1 = st.t1 or time.perf_counter()
\n dur = st.t1 - st.t0
\n utilization = st.utilization or {}
\n it = utilization.get(\"input_tokens\")
\n ot = utilization.get(\"output_tokens\")
\n complete = utilization.get(\"total_tokens\")
\n if complete is None and (it isn't None or ot is just not None):
\n complete = (it or 0) + (ot or 0)
\n tools_used = \", \".be part of(f\"{okay}\u00d7{v}\" for okay, v in st.instruments.objects()) or \"--\"
\n model_label = friendly_model(st.model_raw or opts_model)
\n
\n field = [
\n \"\u250c\u2500 Run Summary \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n f\"\u2502 Model: {model_label}\",
\n f\"\u2502 Tokens: {total if total is not None else '?'}\"
\n + (f\" (in={it if it is not None else '?'} | out={ot if ot is not None else '?'})\"
\n if (it is not None or ot is not None) else \"\"),
\n f\"\u2502 Tools: {tools_used}\",
\n f\"\u2502 Duration: {dur:.1f}s\",
\n \"\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n ]
\n print(GRAY + \"n\".be part of(field) + RESET, file=sys.stderr)<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>6. The Predominant Loop (All-in-One)<\/h4>\nThis begins the app, reads your command, asks the AI, streams the reply, saves \/development<\/code> experiences, and prints the abstract.<\/p>\n\nasync def most important():
\n \"\"\"Setup \u2192 REPL \u2192 parse \u2192 question\/stream \u2192 auto-save \u2192 abstract.\"\"\"
\n st = State()
\n _ensure_dir(REPORTS)
\n
\n opts = ClaudeAgentOptions(
\n mannequin=MODEL,
\n system_prompt=SYS,
\n allowed_tools=[\"WebFetch\", \"WebSearch\"],
\n )
\n
\n print(\"\ud83d\udcc8 TrendSmith nn\" + HELP)
\n
\n async with ClaudeSDKClient(choices=opts) as shopper:
\n whereas True:
\n # Learn enter
\n strive:
\n consumer = enter(\"nYou: \").strip()
\n besides (EOFError, KeyboardInterrupt):
\n print(\"nBye!\")
\n break
\n
\n if not consumer:
\n proceed
\n low = consumer.decrease()
\n
\n # Fundamental instructions
\n if low in {\"\/exit\", \"exit\", \"stop\"}:
\n print(\"Bye!\")
\n break
\n if low in {\"\/assist\", \"assist\"}:
\n print(HELP)
\n proceed
\n
\n # Parse right into a immediate
\n if low.startswith(\"\/development \"):
\n subject = consumer.cut up(\" \", 1)[1].strip().strip('\"')
\n if not subject:
\n print('e.g. \/development \"AI chip startups\"')
\n proceed
\n st.last_cmd, st.last_topic = \"development\", subject
\n immediate = f\"Run a quick development scan for '{subject}' following the output spec.\"
\n elif low.startswith(\"\/scan \"):
\n q = consumer.cut up(\" \", 1)[1].strip()
\n if not q:
\n print('e.g. \/scan \"AI {hardware} information\"')
\n proceed
\n st.last_cmd, st.last_topic = \"scan\", q
\n immediate = f\"Fast scan for '{q}' in underneath 10s (\u22641 WebFetch). Return 5 bullets + TL;DR + sources.\"
\n else:
\n st.last_cmd, st.last_topic = \"free\", None
\n immediate = consumer
\n
\n # Execute request and stream outcomes
\n st.reset()
\n print(f\"{GRAY}\u25b6 Working...{RESET}\")
\n strive:
\n await shopper.question(immediate)
\n besides Exception as e:
\n print(f\"{GRAY}\u274c Question error: {e}{RESET}\")
\n proceed
\n
\n strive:
\n async for m in shopper.receive_response():
\n if isinstance(m, AssistantMessage):
\n st.model_raw = st.model_raw or m.mannequin
\n for b in m.content material:
\n if isinstance(b, TextBlock):
\n st.temporary += b.textual content or \"\"
\n print(b.textual content or \"\", finish=\"\")
\n elif isinstance(b, ToolUseBlock):
\n identify = b.identify or \"Device\"
\n st.instruments[name] = st.instruments.get(identify, 0) + 1
\n print(f\"{GRAY}n\ud83d\udee0 Device: {identify}{RESET}\")
\n elif isinstance(b, ToolResultBlock):
\n go # quiet instrument payloads
\n elif isinstance(m, ResultMessage):
\n st.utilization = m.utilization or {}
\n st.value = m.total_cost_usd
\n besides Exception as e:
\n print(f\"{GRAY}n\u26a0 Stream error: {e}{RESET}\")
\n
\n # Auto-save development briefs and present the abstract
\n if st.last_cmd == \"development\" and st.temporary.strip():
\n strive:
\n saved_path = save_report(st.last_topic or \"development\", st.temporary)
\n print(f\"n{GRAY}\u2705 Auto-saved \u2192 {saved_path}{RESET}\")
\n besides Exception as e:
\n print(f\"{GRAY}\u26a0 Save error: {e}{RESET}\")
\n
\n st.t1 = time.perf_counter()
\n usage_footer(st, opts.mannequin)
\n
\nif __name__ == \"__main__\":
\n asyncio.run(most important())<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
#\u00a0<\/span>Testing the TrendSmith Utility<\/h2>\n\u00a0
We are going to now take a look at the app by operating the Python file. Here’s a fast recap on learn how to use the CLI software:<\/p>\n
\n- \/development “
“<\/topic><\/strong> \u2192 temporary multi-source scan, auto-saved to experiences\/_.md<\/ts><\/topic><\/code>.\n<\/li>\n- \/scan “
“<\/query><\/strong> \u2192 one-page fast scan (\u22641 WebFetch), prints solely.\n<\/li>\n- \/assist<\/strong> \u2192 exhibits instructions.\n<\/li>\n
- \/exit<\/strong> \u2192 quits.\n<\/li>\n<\/ul>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe have now used the \/development<\/code> choice to seek for AI chip startups.<\/p>\n\n\/development \"AI chip startups\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
In consequence, the app has used varied search and net scraping instruments to collect info from completely different web sites.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nFinally, it has supplied the total response, auto-saved the report within the markdown file, and generated the utilization abstract. It value us $0.136.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nHere’s a preview of the saved Markdown report on the AI Chips Startups.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe are going to now take a look at the scanning choice and generate a abstract in regards to the subject utilizing an internet search.<\/p>\n
\u00a0<\/p>\n
It makes use of a easy net search and fetch instrument to generate a brief abstract on the subject.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center><\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Last Ideas<\/h2>\n\u00a0
This app ran easily, and dealing with the Claude Agent SDK<\/strong> was genuinely enjoyable. If you’re already on the Claude Code<\/strong> plan, I extremely suggest attempting it to rework your day-to-day terminal workflow into dependable, repeatable agentic CLIs<\/strong>.<\/p>\nUse it to:<\/p>\n
\n- Automate widespread dev duties (debug, take a look at, deploy).\n<\/li>\n
- Script easy analytics or ops routines.\n<\/li>\n
- Bundle your circulation right into a reusable, shareable instrument.\n<\/li>\n<\/ul>\n
The SDK is an effective match for professionals who need stability<\/strong>, reproducibility<\/strong>, and low glue-code overhead<\/strong>. And sure, you may even ask Claude Code that will help you construct the agentic software itself with the SDK.
\u00a0
\u00a0<\/p>\nAbid Ali Awan<\/a><\/strong><\/b><\/a> (@1abidaliawan<\/a>) is a licensed knowledge scientist skilled who loves constructing machine studying fashions. Presently, he’s specializing in content material creation and writing technical blogs on machine studying and knowledge science applied sciences. Abid holds a Grasp’s diploma in expertise administration and a bachelor’s diploma in telecommunication engineering. His imaginative and prescient is to construct an AI product utilizing a graph neural community for college students combating psychological sickness.<\/p>\n<\/p><\/div>\n\n","protected":false},"excerpt":{"rendered":"Picture by Writer \u00a0 #\u00a0Introduction \u00a0Uninterested in duct-taping scripts, instruments, and prompts collectively? The Claude Agent SDK enables you to flip your Claude Code \u201cplan \u2192 construct \u2192 run\u201d workflow into actual, programmable brokers, so you may automate duties, wire up instruments, and ship command line interface (CLI) apps with out tons of glue code. […]<\/p>\n","protected":false},"author":2,"featured_media":9222,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[55],"tags":[75,458,721,2296],"class_list":["post-9220","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-machine-learning","tag-agent","tag-claude","tag-sdk","tag-started"],"_links":{"self":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9220"}],"version-history":[{"count":1,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions"}],"predecessor-version":[{"id":9221,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions\/9221"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/media\/9222"}],"wp:attachment":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9220"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9220"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9220"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
\u00a0<\/p>\n
\/\/\u00a0<\/span>2. Set up Claude Code CLI<\/h4>\nWe are going to set up the Claude Code CLI on Home windows by typing the next command in PowerShell:<\/p>\n
\nirm https:\/\/claude.ai\/set up.ps1 | iex<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
Then add this path to your system surroundings:<\/p>\n
\u00a0<\/p>\n
Restart PowerShell and take a look at:<\/p>\n
\u00a0<\/p>\n
For different platforms, think about using the npm bundle supervisor:<\/p>\n
\nnpm i -g @anthropic-ai\/claude-code<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
After set up, kind claude<\/code> in your terminal to check in.<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Set up the Claude Agent SDK (Python)<\/h4>\nSet up the Claude Agent Python SDK utilizing the pip bundle supervisor.<\/p>\n
\npip set up claude-agent-sdk<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
When you get a CLINotFoundError<\/code>, make sure the Claude CLI is accurately put in and included in your PATH.<\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Constructing a Multi-Device App with the Claude Agent SDK<\/h2>\n\u00a0
On this part, we are going to construct the TrendSmith<\/strong> software, which tracks dwell market traits throughout varied industries, together with startups, AI, finance, and sustainability.<\/p>\nIt combines Claude Sonnet 4.5<\/strong>, WebSearch<\/strong>, WebFetch<\/strong>, and native storage<\/strong> instruments right into a single multi-agent system.<\/p>\nCreate the Python file trend_smith.py<\/code> and add the next code to it:<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>1. Imports & Fundamental Settings<\/h4>\nThis masses Python libraries, the Claude Agent SDK sorts, a tiny assist menu, the mannequin identify, and smooth grey textual content styling for standing traces.<\/p>\n
\nimport asyncio
\nimport os
\nimport re
\nimport sys
\nimport time
\nfrom datetime import datetime
\nfrom pathlib import Path
\n
\nfrom claude_agent_sdk import (
\n AssistantMessage,
\n ClaudeAgentOptions,
\n ClaudeSDKClient,
\n ResultMessage,
\n TextBlock,
\n ToolResultBlock,
\n ToolUseBlock,
\n)
\n
\nHELP = \"\"\"Instructions:
\n\/development Fast multi-source scan (auto-saves markdown)
\n\/scan Quick one-page scan
\n\/assist \/exit Assist \/ Stop
\n\"\"\"
\n
\nMODEL = os.getenv(\"CLAUDE_MODEL\", \"sonnet\") # e.g. \"sonnet-4.5\"
\nGRAY = \"\u000033[90m\"
\nRESET = \"\u000033[0m\"<\/query><\/topic><\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>2. System Prompt & Report Destination<\/h4>\nThis sets the \u201chouse rules\u201d for answers (fast, compact, consistent sections) and chooses a reports\/<\/code> folder next to your script for saved briefs.<\/p>\n\nSYS = \"\"\"You are TrendSmith, a fast, concise trend researcher.
\n- Finish quickly (~20 s).
\n- For \/trend: \u22641 WebSearch + \u22642 WebFetch from distinct domains.
\n- For \/scan: \u22641 WebFetch only.
\nReturn for \/trend:
\n TL;DR (1 line)
\n 3-5 Signals (short bullets)
\n Key Players, Risks, 30\/90-day Watchlist
\n Sources (markdown: **Title** -- URL)
\nReturn for \/scan: 5 bullets + TL;DR + Sources.
\nAfter finishing \/trend, the client will auto-save your full brief.
\n\"\"\"
\n
\nBASE = Path(__file__).parent
\nREPORTS = BASE \/ \"reports\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Saving Files Safely<\/h4>\nThese helpers make filenames safe, create folders if needed, and always try a home-folder fallback so your report still gets saved.<\/p>\n
\ndef _ts():
\n return datetime.now().strftime(\"%Y%m%d_%H%M\")
\n
\ndef _sanitize(s: str):
\n return re.sub(r\"[^w-.]+\", \"_\", s).strip(\"_\") or \"untitled\"
\n
\ndef _ensure_dir(p: Path):
\n strive:
\n p.mkdir(dad and mom=True, exist_ok=True)
\n besides Exception:
\n go
\n
\ndef _safe_write(path: Path, textual content: str) -> Path:
\n \"\"\"Write textual content to path; if listing\/permission fails, fall again to ~\/TrendSmith\/experiences.\"\"\"
\n strive:
\n _ensure_dir(path.mum or dad)
\n path.write_text(textual content, encoding=\"utf-8\")
\n return path
\n besides Exception:
\n home_reports = Path.dwelling() \/ \"TrendSmith\"https:\/\/www.kdnuggets.com\/\"experiences\"
\n _ensure_dir(home_reports)
\n fb = home_reports \/ path.identify
\n fb.write_text(textual content, encoding=\"utf-8\")
\n return fb
\n
\ndef save_report(subject: str, textual content: str) -> Path:
\n filename = f\"{_sanitize(subject)}_{_ts()}.md\"
\n goal = REPORTS \/ filename
\n return _safe_write(goal, textual content.strip() + \"n\")<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>4. Monitoring Every Run<\/h4>\nThis retains what you want for one request: streamed textual content, mannequin, instrument counts, token utilization, and timing, then resets cleanly earlier than the following request.<\/p>\n
\nclass State:
\n def __init__(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.last_cmd = None
\n self.last_topic = None
\n self.instruments = {}
\n self.t0 = 0.0
\n self.t1 = 0.0
\n
\n def reset(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.instruments = {}
\n self.t0 = time.perf_counter()
\n self.t1 = 0.0
\n
\ndef friendly_model(identify: str | None) -> str:
\n if not identify:
\n return MODEL
\n n = (identify or \"\").decrease()
\n if \"sonnet-4-5\" in n or \"sonnet_4_5\" in n:
\n return \"Claude 4.5 Sonnet\"
\n if \"sonnet\" in n:
\n return \"Claude Sonnet\"
\n if \"haiku\" in n:
\n return \"Claude Haiku\"
\n if \"opus\" in n:
\n return \"Claude Opus\"
\n return identify or \"Unknown\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>5. Quick Run Abstract<\/h4>\nThis prints a neat grey field to indicate the mannequin, tokens, instrument utilization, and period, with out mixing into your streamed content material.<\/p>\n
\ndef usage_footer(st: State, opts_model: str):
\n st.t1 = st.t1 or time.perf_counter()
\n dur = st.t1 - st.t0
\n utilization = st.utilization or {}
\n it = utilization.get(\"input_tokens\")
\n ot = utilization.get(\"output_tokens\")
\n complete = utilization.get(\"total_tokens\")
\n if complete is None and (it isn't None or ot is just not None):
\n complete = (it or 0) + (ot or 0)
\n tools_used = \", \".be part of(f\"{okay}\u00d7{v}\" for okay, v in st.instruments.objects()) or \"--\"
\n model_label = friendly_model(st.model_raw or opts_model)
\n
\n field = [
\n \"\u250c\u2500 Run Summary \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n f\"\u2502 Model: {model_label}\",
\n f\"\u2502 Tokens: {total if total is not None else '?'}\"
\n + (f\" (in={it if it is not None else '?'} | out={ot if ot is not None else '?'})\"
\n if (it is not None or ot is not None) else \"\"),
\n f\"\u2502 Tools: {tools_used}\",
\n f\"\u2502 Duration: {dur:.1f}s\",
\n \"\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n ]
\n print(GRAY + \"n\".be part of(field) + RESET, file=sys.stderr)<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>6. The Predominant Loop (All-in-One)<\/h4>\nThis begins the app, reads your command, asks the AI, streams the reply, saves \/development<\/code> experiences, and prints the abstract.<\/p>\n\nasync def most important():
\n \"\"\"Setup \u2192 REPL \u2192 parse \u2192 question\/stream \u2192 auto-save \u2192 abstract.\"\"\"
\n st = State()
\n _ensure_dir(REPORTS)
\n
\n opts = ClaudeAgentOptions(
\n mannequin=MODEL,
\n system_prompt=SYS,
\n allowed_tools=[\"WebFetch\", \"WebSearch\"],
\n )
\n
\n print(\"\ud83d\udcc8 TrendSmith nn\" + HELP)
\n
\n async with ClaudeSDKClient(choices=opts) as shopper:
\n whereas True:
\n # Learn enter
\n strive:
\n consumer = enter(\"nYou: \").strip()
\n besides (EOFError, KeyboardInterrupt):
\n print(\"nBye!\")
\n break
\n
\n if not consumer:
\n proceed
\n low = consumer.decrease()
\n
\n # Fundamental instructions
\n if low in {\"\/exit\", \"exit\", \"stop\"}:
\n print(\"Bye!\")
\n break
\n if low in {\"\/assist\", \"assist\"}:
\n print(HELP)
\n proceed
\n
\n # Parse right into a immediate
\n if low.startswith(\"\/development \"):
\n subject = consumer.cut up(\" \", 1)[1].strip().strip('\"')
\n if not subject:
\n print('e.g. \/development \"AI chip startups\"')
\n proceed
\n st.last_cmd, st.last_topic = \"development\", subject
\n immediate = f\"Run a quick development scan for '{subject}' following the output spec.\"
\n elif low.startswith(\"\/scan \"):
\n q = consumer.cut up(\" \", 1)[1].strip()
\n if not q:
\n print('e.g. \/scan \"AI {hardware} information\"')
\n proceed
\n st.last_cmd, st.last_topic = \"scan\", q
\n immediate = f\"Fast scan for '{q}' in underneath 10s (\u22641 WebFetch). Return 5 bullets + TL;DR + sources.\"
\n else:
\n st.last_cmd, st.last_topic = \"free\", None
\n immediate = consumer
\n
\n # Execute request and stream outcomes
\n st.reset()
\n print(f\"{GRAY}\u25b6 Working...{RESET}\")
\n strive:
\n await shopper.question(immediate)
\n besides Exception as e:
\n print(f\"{GRAY}\u274c Question error: {e}{RESET}\")
\n proceed
\n
\n strive:
\n async for m in shopper.receive_response():
\n if isinstance(m, AssistantMessage):
\n st.model_raw = st.model_raw or m.mannequin
\n for b in m.content material:
\n if isinstance(b, TextBlock):
\n st.temporary += b.textual content or \"\"
\n print(b.textual content or \"\", finish=\"\")
\n elif isinstance(b, ToolUseBlock):
\n identify = b.identify or \"Device\"
\n st.instruments[name] = st.instruments.get(identify, 0) + 1
\n print(f\"{GRAY}n\ud83d\udee0 Device: {identify}{RESET}\")
\n elif isinstance(b, ToolResultBlock):
\n go # quiet instrument payloads
\n elif isinstance(m, ResultMessage):
\n st.utilization = m.utilization or {}
\n st.value = m.total_cost_usd
\n besides Exception as e:
\n print(f\"{GRAY}n\u26a0 Stream error: {e}{RESET}\")
\n
\n # Auto-save development briefs and present the abstract
\n if st.last_cmd == \"development\" and st.temporary.strip():
\n strive:
\n saved_path = save_report(st.last_topic or \"development\", st.temporary)
\n print(f\"n{GRAY}\u2705 Auto-saved \u2192 {saved_path}{RESET}\")
\n besides Exception as e:
\n print(f\"{GRAY}\u26a0 Save error: {e}{RESET}\")
\n
\n st.t1 = time.perf_counter()
\n usage_footer(st, opts.mannequin)
\n
\nif __name__ == \"__main__\":
\n asyncio.run(most important())<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
#\u00a0<\/span>Testing the TrendSmith Utility<\/h2>\n\u00a0
We are going to now take a look at the app by operating the Python file. Here’s a fast recap on learn how to use the CLI software:<\/p>\n
\n- \/development “
“<\/topic><\/strong> \u2192 temporary multi-source scan, auto-saved to experiences\/_.md<\/ts><\/topic><\/code>.\n<\/li>\n- \/scan “
“<\/query><\/strong> \u2192 one-page fast scan (\u22641 WebFetch), prints solely.\n<\/li>\n- \/assist<\/strong> \u2192 exhibits instructions.\n<\/li>\n
- \/exit<\/strong> \u2192 quits.\n<\/li>\n<\/ul>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe have now used the \/development<\/code> choice to seek for AI chip startups.<\/p>\n\n\/development \"AI chip startups\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
In consequence, the app has used varied search and net scraping instruments to collect info from completely different web sites.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nFinally, it has supplied the total response, auto-saved the report within the markdown file, and generated the utilization abstract. It value us $0.136.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nHere’s a preview of the saved Markdown report on the AI Chips Startups.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe are going to now take a look at the scanning choice and generate a abstract in regards to the subject utilizing an internet search.<\/p>\n
\u00a0<\/p>\n
It makes use of a easy net search and fetch instrument to generate a brief abstract on the subject.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center><\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Last Ideas<\/h2>\n\u00a0
This app ran easily, and dealing with the Claude Agent SDK<\/strong> was genuinely enjoyable. If you’re already on the Claude Code<\/strong> plan, I extremely suggest attempting it to rework your day-to-day terminal workflow into dependable, repeatable agentic CLIs<\/strong>.<\/p>\nUse it to:<\/p>\n
\n- Automate widespread dev duties (debug, take a look at, deploy).\n<\/li>\n
- Script easy analytics or ops routines.\n<\/li>\n
- Bundle your circulation right into a reusable, shareable instrument.\n<\/li>\n<\/ul>\n
The SDK is an effective match for professionals who need stability<\/strong>, reproducibility<\/strong>, and low glue-code overhead<\/strong>. And sure, you may even ask Claude Code that will help you construct the agentic software itself with the SDK.
\u00a0
\u00a0<\/p>\nAbid Ali Awan<\/a><\/strong><\/b><\/a> (@1abidaliawan<\/a>) is a licensed knowledge scientist skilled who loves constructing machine studying fashions. Presently, he’s specializing in content material creation and writing technical blogs on machine studying and knowledge science applied sciences. Abid holds a Grasp’s diploma in expertise administration and a bachelor’s diploma in telecommunication engineering. His imaginative and prescient is to construct an AI product utilizing a graph neural community for college students combating psychological sickness.<\/p>\n<\/p><\/div>\n\n","protected":false},"excerpt":{"rendered":"Picture by Writer \u00a0 #\u00a0Introduction \u00a0Uninterested in duct-taping scripts, instruments, and prompts collectively? The Claude Agent SDK enables you to flip your Claude Code \u201cplan \u2192 construct \u2192 run\u201d workflow into actual, programmable brokers, so you may automate duties, wire up instruments, and ship command line interface (CLI) apps with out tons of glue code. […]<\/p>\n","protected":false},"author":2,"featured_media":9222,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[55],"tags":[75,458,721,2296],"class_list":["post-9220","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-machine-learning","tag-agent","tag-claude","tag-sdk","tag-started"],"_links":{"self":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9220"}],"version-history":[{"count":1,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions"}],"predecessor-version":[{"id":9221,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions\/9221"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/media\/9222"}],"wp:attachment":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9220"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9220"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9220"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
irm https:\/\/claude.ai\/set up.ps1 | iex<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
Then add this path to your system surroundings:<\/p>\n
\u00a0<\/p>\n
Restart PowerShell and take a look at:<\/p>\n
\u00a0<\/p>\n
For different platforms, think about using the npm bundle supervisor:<\/p>\n
\nnpm i -g @anthropic-ai\/claude-code<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
After set up, kind claude<\/code> in your terminal to check in.<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Set up the Claude Agent SDK (Python)<\/h4>\nSet up the Claude Agent Python SDK utilizing the pip bundle supervisor.<\/p>\n
\npip set up claude-agent-sdk<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
When you get a CLINotFoundError<\/code>, make sure the Claude CLI is accurately put in and included in your PATH.<\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Constructing a Multi-Device App with the Claude Agent SDK<\/h2>\n\u00a0
On this part, we are going to construct the TrendSmith<\/strong> software, which tracks dwell market traits throughout varied industries, together with startups, AI, finance, and sustainability.<\/p>\nIt combines Claude Sonnet 4.5<\/strong>, WebSearch<\/strong>, WebFetch<\/strong>, and native storage<\/strong> instruments right into a single multi-agent system.<\/p>\nCreate the Python file trend_smith.py<\/code> and add the next code to it:<\/p>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>1. Imports & Fundamental Settings<\/h4>\nThis masses Python libraries, the Claude Agent SDK sorts, a tiny assist menu, the mannequin identify, and smooth grey textual content styling for standing traces.<\/p>\n
\nimport asyncio
\nimport os
\nimport re
\nimport sys
\nimport time
\nfrom datetime import datetime
\nfrom pathlib import Path
\n
\nfrom claude_agent_sdk import (
\n AssistantMessage,
\n ClaudeAgentOptions,
\n ClaudeSDKClient,
\n ResultMessage,
\n TextBlock,
\n ToolResultBlock,
\n ToolUseBlock,
\n)
\n
\nHELP = \"\"\"Instructions:
\n\/development Fast multi-source scan (auto-saves markdown)
\n\/scan Quick one-page scan
\n\/assist \/exit Assist \/ Stop
\n\"\"\"
\n
\nMODEL = os.getenv(\"CLAUDE_MODEL\", \"sonnet\") # e.g. \"sonnet-4.5\"
\nGRAY = \"\u000033[90m\"
\nRESET = \"\u000033[0m\"<\/query><\/topic><\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>2. System Prompt & Report Destination<\/h4>\nThis sets the \u201chouse rules\u201d for answers (fast, compact, consistent sections) and chooses a reports\/<\/code> folder next to your script for saved briefs.<\/p>\n\nSYS = \"\"\"You are TrendSmith, a fast, concise trend researcher.
\n- Finish quickly (~20 s).
\n- For \/trend: \u22641 WebSearch + \u22642 WebFetch from distinct domains.
\n- For \/scan: \u22641 WebFetch only.
\nReturn for \/trend:
\n TL;DR (1 line)
\n 3-5 Signals (short bullets)
\n Key Players, Risks, 30\/90-day Watchlist
\n Sources (markdown: **Title** -- URL)
\nReturn for \/scan: 5 bullets + TL;DR + Sources.
\nAfter finishing \/trend, the client will auto-save your full brief.
\n\"\"\"
\n
\nBASE = Path(__file__).parent
\nREPORTS = BASE \/ \"reports\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>3. Saving Files Safely<\/h4>\nThese helpers make filenames safe, create folders if needed, and always try a home-folder fallback so your report still gets saved.<\/p>\n
\ndef _ts():
\n return datetime.now().strftime(\"%Y%m%d_%H%M\")
\n
\ndef _sanitize(s: str):
\n return re.sub(r\"[^w-.]+\", \"_\", s).strip(\"_\") or \"untitled\"
\n
\ndef _ensure_dir(p: Path):
\n strive:
\n p.mkdir(dad and mom=True, exist_ok=True)
\n besides Exception:
\n go
\n
\ndef _safe_write(path: Path, textual content: str) -> Path:
\n \"\"\"Write textual content to path; if listing\/permission fails, fall again to ~\/TrendSmith\/experiences.\"\"\"
\n strive:
\n _ensure_dir(path.mum or dad)
\n path.write_text(textual content, encoding=\"utf-8\")
\n return path
\n besides Exception:
\n home_reports = Path.dwelling() \/ \"TrendSmith\"https:\/\/www.kdnuggets.com\/\"experiences\"
\n _ensure_dir(home_reports)
\n fb = home_reports \/ path.identify
\n fb.write_text(textual content, encoding=\"utf-8\")
\n return fb
\n
\ndef save_report(subject: str, textual content: str) -> Path:
\n filename = f\"{_sanitize(subject)}_{_ts()}.md\"
\n goal = REPORTS \/ filename
\n return _safe_write(goal, textual content.strip() + \"n\")<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>4. Monitoring Every Run<\/h4>\nThis retains what you want for one request: streamed textual content, mannequin, instrument counts, token utilization, and timing, then resets cleanly earlier than the following request.<\/p>\n
\nclass State:
\n def __init__(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.last_cmd = None
\n self.last_topic = None
\n self.instruments = {}
\n self.t0 = 0.0
\n self.t1 = 0.0
\n
\n def reset(self):
\n self.temporary = \"\"
\n self.model_raw = None
\n self.utilization = {}
\n self.value = None
\n self.instruments = {}
\n self.t0 = time.perf_counter()
\n self.t1 = 0.0
\n
\ndef friendly_model(identify: str | None) -> str:
\n if not identify:
\n return MODEL
\n n = (identify or \"\").decrease()
\n if \"sonnet-4-5\" in n or \"sonnet_4_5\" in n:
\n return \"Claude 4.5 Sonnet\"
\n if \"sonnet\" in n:
\n return \"Claude Sonnet\"
\n if \"haiku\" in n:
\n return \"Claude Haiku\"
\n if \"opus\" in n:
\n return \"Claude Opus\"
\n return identify or \"Unknown\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>5. Quick Run Abstract<\/h4>\nThis prints a neat grey field to indicate the mannequin, tokens, instrument utilization, and period, with out mixing into your streamed content material.<\/p>\n
\ndef usage_footer(st: State, opts_model: str):
\n st.t1 = st.t1 or time.perf_counter()
\n dur = st.t1 - st.t0
\n utilization = st.utilization or {}
\n it = utilization.get(\"input_tokens\")
\n ot = utilization.get(\"output_tokens\")
\n complete = utilization.get(\"total_tokens\")
\n if complete is None and (it isn't None or ot is just not None):
\n complete = (it or 0) + (ot or 0)
\n tools_used = \", \".be part of(f\"{okay}\u00d7{v}\" for okay, v in st.instruments.objects()) or \"--\"
\n model_label = friendly_model(st.model_raw or opts_model)
\n
\n field = [
\n \"\u250c\u2500 Run Summary \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n f\"\u2502 Model: {model_label}\",
\n f\"\u2502 Tokens: {total if total is not None else '?'}\"
\n + (f\" (in={it if it is not None else '?'} | out={ot if ot is not None else '?'})\"
\n if (it is not None or ot is not None) else \"\"),
\n f\"\u2502 Tools: {tools_used}\",
\n f\"\u2502 Duration: {dur:.1f}s\",
\n \"\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\",
\n ]
\n print(GRAY + \"n\".be part of(field) + RESET, file=sys.stderr)<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
\/\/\u00a0<\/span>6. The Predominant Loop (All-in-One)<\/h4>\nThis begins the app, reads your command, asks the AI, streams the reply, saves \/development<\/code> experiences, and prints the abstract.<\/p>\n\nasync def most important():
\n \"\"\"Setup \u2192 REPL \u2192 parse \u2192 question\/stream \u2192 auto-save \u2192 abstract.\"\"\"
\n st = State()
\n _ensure_dir(REPORTS)
\n
\n opts = ClaudeAgentOptions(
\n mannequin=MODEL,
\n system_prompt=SYS,
\n allowed_tools=[\"WebFetch\", \"WebSearch\"],
\n )
\n
\n print(\"\ud83d\udcc8 TrendSmith nn\" + HELP)
\n
\n async with ClaudeSDKClient(choices=opts) as shopper:
\n whereas True:
\n # Learn enter
\n strive:
\n consumer = enter(\"nYou: \").strip()
\n besides (EOFError, KeyboardInterrupt):
\n print(\"nBye!\")
\n break
\n
\n if not consumer:
\n proceed
\n low = consumer.decrease()
\n
\n # Fundamental instructions
\n if low in {\"\/exit\", \"exit\", \"stop\"}:
\n print(\"Bye!\")
\n break
\n if low in {\"\/assist\", \"assist\"}:
\n print(HELP)
\n proceed
\n
\n # Parse right into a immediate
\n if low.startswith(\"\/development \"):
\n subject = consumer.cut up(\" \", 1)[1].strip().strip('\"')
\n if not subject:
\n print('e.g. \/development \"AI chip startups\"')
\n proceed
\n st.last_cmd, st.last_topic = \"development\", subject
\n immediate = f\"Run a quick development scan for '{subject}' following the output spec.\"
\n elif low.startswith(\"\/scan \"):
\n q = consumer.cut up(\" \", 1)[1].strip()
\n if not q:
\n print('e.g. \/scan \"AI {hardware} information\"')
\n proceed
\n st.last_cmd, st.last_topic = \"scan\", q
\n immediate = f\"Fast scan for '{q}' in underneath 10s (\u22641 WebFetch). Return 5 bullets + TL;DR + sources.\"
\n else:
\n st.last_cmd, st.last_topic = \"free\", None
\n immediate = consumer
\n
\n # Execute request and stream outcomes
\n st.reset()
\n print(f\"{GRAY}\u25b6 Working...{RESET}\")
\n strive:
\n await shopper.question(immediate)
\n besides Exception as e:
\n print(f\"{GRAY}\u274c Question error: {e}{RESET}\")
\n proceed
\n
\n strive:
\n async for m in shopper.receive_response():
\n if isinstance(m, AssistantMessage):
\n st.model_raw = st.model_raw or m.mannequin
\n for b in m.content material:
\n if isinstance(b, TextBlock):
\n st.temporary += b.textual content or \"\"
\n print(b.textual content or \"\", finish=\"\")
\n elif isinstance(b, ToolUseBlock):
\n identify = b.identify or \"Device\"
\n st.instruments[name] = st.instruments.get(identify, 0) + 1
\n print(f\"{GRAY}n\ud83d\udee0 Device: {identify}{RESET}\")
\n elif isinstance(b, ToolResultBlock):
\n go # quiet instrument payloads
\n elif isinstance(m, ResultMessage):
\n st.utilization = m.utilization or {}
\n st.value = m.total_cost_usd
\n besides Exception as e:
\n print(f\"{GRAY}n\u26a0 Stream error: {e}{RESET}\")
\n
\n # Auto-save development briefs and present the abstract
\n if st.last_cmd == \"development\" and st.temporary.strip():
\n strive:
\n saved_path = save_report(st.last_topic or \"development\", st.temporary)
\n print(f\"n{GRAY}\u2705 Auto-saved \u2192 {saved_path}{RESET}\")
\n besides Exception as e:
\n print(f\"{GRAY}\u26a0 Save error: {e}{RESET}\")
\n
\n st.t1 = time.perf_counter()
\n usage_footer(st, opts.mannequin)
\n
\nif __name__ == \"__main__\":
\n asyncio.run(most important())<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
#\u00a0<\/span>Testing the TrendSmith Utility<\/h2>\n\u00a0
We are going to now take a look at the app by operating the Python file. Here’s a fast recap on learn how to use the CLI software:<\/p>\n
\n- \/development “
“<\/topic><\/strong> \u2192 temporary multi-source scan, auto-saved to experiences\/_.md<\/ts><\/topic><\/code>.\n<\/li>\n- \/scan “
“<\/query><\/strong> \u2192 one-page fast scan (\u22641 WebFetch), prints solely.\n<\/li>\n- \/assist<\/strong> \u2192 exhibits instructions.\n<\/li>\n
- \/exit<\/strong> \u2192 quits.\n<\/li>\n<\/ul>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe have now used the \/development<\/code> choice to seek for AI chip startups.<\/p>\n\n\/development \"AI chip startups\"<\/code><\/pre>\n<\/div>\n\u00a0<\/p>\n
In consequence, the app has used varied search and net scraping instruments to collect info from completely different web sites.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nFinally, it has supplied the total response, auto-saved the report within the markdown file, and generated the utilization abstract. It value us $0.136.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nHere’s a preview of the saved Markdown report on the AI Chips Startups.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center>
\n\u00a0<\/p>\nWe are going to now take a look at the scanning choice and generate a abstract in regards to the subject utilizing an internet search.<\/p>\n
\u00a0<\/p>\n
It makes use of a easy net search and fetch instrument to generate a brief abstract on the subject.<\/p>\n
\u00a0<\/p>\n
Picture by Writer<\/span><\/center><\/p>\n\u00a0<\/p>\n
#\u00a0<\/span>Last Ideas<\/h2>\n\u00a0
This app ran easily, and dealing with the Claude Agent SDK<\/strong> was genuinely enjoyable. If you’re already on the Claude Code<\/strong> plan, I extremely suggest attempting it to rework your day-to-day terminal workflow into dependable, repeatable agentic CLIs<\/strong>.<\/p>\nUse it to:<\/p>\n
\n- Automate widespread dev duties (debug, take a look at, deploy).\n<\/li>\n
- Script easy analytics or ops routines.\n<\/li>\n
- Bundle your circulation right into a reusable, shareable instrument.\n<\/li>\n<\/ul>\n
The SDK is an effective match for professionals who need stability<\/strong>, reproducibility<\/strong>, and low glue-code overhead<\/strong>. And sure, you may even ask Claude Code that will help you construct the agentic software itself with the SDK.
\u00a0
\u00a0<\/p>\nAbid Ali Awan<\/a><\/strong><\/b><\/a> (@1abidaliawan<\/a>) is a licensed knowledge scientist skilled who loves constructing machine studying fashions. Presently, he’s specializing in content material creation and writing technical blogs on machine studying and knowledge science applied sciences. Abid holds a Grasp’s diploma in expertise administration and a bachelor’s diploma in telecommunication engineering. His imaginative and prescient is to construct an AI product utilizing a graph neural community for college students combating psychological sickness.<\/p>\n<\/p><\/div>\n\n","protected":false},"excerpt":{"rendered":"Picture by Writer \u00a0 #\u00a0Introduction \u00a0Uninterested in duct-taping scripts, instruments, and prompts collectively? The Claude Agent SDK enables you to flip your Claude Code \u201cplan \u2192 construct \u2192 run\u201d workflow into actual, programmable brokers, so you may automate duties, wire up instruments, and ship command line interface (CLI) apps with out tons of glue code. […]<\/p>\n","protected":false},"author":2,"featured_media":9222,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[55],"tags":[75,458,721,2296],"class_list":["post-9220","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-machine-learning","tag-agent","tag-claude","tag-sdk","tag-started"],"_links":{"self":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9220"}],"version-history":[{"count":1,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions"}],"predecessor-version":[{"id":9221,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/9220\/revisions\/9221"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/media\/9222"}],"wp:attachment":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9220"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9220"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9220"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}