Your next AI side project — the one that automates your inbox, crunches reports, or drafts code — won’t die in framework hell.
Selectools hands you the keys to real AI agents, the kind that hum along in production without LangChain’s hidden gotchas or pricey tracing dashboards. It’s like swapping a minivan full of junk drawers for a sleek electric bike: lighter, faster, and you actually control where it goes.
And here’s the thrill — this isn’t vaporware from a VC-fueled startup. John Nichev, scarred by enterprise battles, built it because existing tools crumbled under real workloads. Picture your agents pausing for your nod, resuming exactly where they left off, no restarts. Or chaining three brains — planner, writer, reviewer — in one Python line. That’s not hype; it’s shipping.
“Tool calling that just works. Define a function, the LLM calls it. No adapter layers, no schema gymnastics. Works the same across OpenAI, Anthropic, Gemini, and Ollama.”
Boom. No more wrestling LCEL’s | operator that nukes your debugger. Selectools keeps it Pythonic: functions pipe with |, traces spill everything locally (tokens, timings, costs), guards block PII or injections out of the box.
Why Does Selectools Feel Like Magic?
Think back to 2005. Rails burst onto the scene promising convention over configuration, and suddenly web apps weren’t boilerplate marathons. Selectools pulls the same trick for AI agents — but bolder.
You define tools as plain functions. LLMs call them smoothly across models. Want orchestration? Python if/else, not some Pregel-inspired graph DSL that demands recompiles. Human-in-the-loop? Yield an interrupt, resume precisely — async def review(state): analysis = await expensive_work(state); decision = yield InterruptRequest(…). No re-running that GPU-melter.
Deployment? selectools serve agent.yaml spins up HTTP, SSE streaming, even a playground. Zero FastAPI fiddling. It’s the deploy button every dev dreams of, minus the vendor lock.
But wait — the visual builder. Drag nodes in your browser, wire edges, tweak models/tools, export YAML or Python. Zero install, on GitHub Pages. That’s democratizing agent design like Figma did for UI.
One line deploys your empire.
result = AgentGraph.chain(planner, writer, reviewer).run(“Write a blog post”)
Or pipe text flows: pipeline = summarize | translate | format. Pure poetry.
Why Real Builders Hate LangChain (And Love This)
LangChain’s a beast — 5 packages for basics, paid LangSmith to debug your code, graphs restarting on pauses. Fine for prototypes. Hell for production.
Selectools? 4,612 tests, 95% coverage, Python 3.9-3.13. Pre-launch audit squashed 9 critical bugs via 5-agent hunts. 152 model defs with pricing. 50 evaluators baked in. Apache-2.0. Smaller footprint, zero SaaS.
My unique take: This echoes jQuery’s rise against browser DOM soup. Back then, devs begged for sane abstractions. Today, AI frameworks bloated into DSLs and platforms. Selectools bets on Python itself as the agent orchestrator — bold prediction: it’ll spawn a ecosystem of pip-install plugins, becoming the lightweight standard while giants chase enterprise.
Sure, if you crave 50 integrations now, stick with LangChain. But for staying nimble? pip install selectools.
Is Selectools Ready to Replace Your Agent Stack?
Short answer: For most indie and startup work, yes. It supports OpenAI, Anthropic, Gemini, Ollama — your faves. Traces rival LangSmith, free. Guards ship-ready.
Cookbook’s packed with 76 examples. Docs? 44 interactive modules, runnable, with stability badges. It’s thoughtful engineering, not slapdash.
The catch? Young community. No massive plugin zoo yet. But that’s the opportunity — contribute, fork, fly.
What if agents were this easy from day one?
We’re in the garage-band phase of AI dev tools. Selectools feels like the Fender Strat that launched rock ‘n’ roll — simple, extensible, lets talent shine without amp feedback.
How Selectools Changes Everything for You
Freelancer juggling client automations? Chain summarizer to email drafter, deploy to Vercel-ish endpoint. Done.
Team building customer-facing bots? Guards enforce compliance, traces audit costs — all local.
Hobbyist tinkering? Visual builder prototypes in browser, export, iterate.
It’s not just a library; it’s the platform shift where AI agents become composable like React components. No more “connect LLM to tools and pray.” You build worlds.
🧬 Related Insights
- Read more: 90 Minutes to a Telegram AI Sidekick with OpenClaw: Clever Hack or Fancy Wrapper?
- Read more: Solo Dev’s React-Firebase ERP: Bold Bets That Paid Off
Frequently Asked Questions
What is selectools?
Selectools is a lightweight Python library for building, tracing, guarding, and deploying AI agents with smoothly tool calling across major LLMs—no bloat, no paid services.
How do I get started with selectools?
pip install selectools, then define functions as tools, chain with AgentGraph or |, and serve with one command. Check https://selectools.dev for docs and cookbook.
Is selectools better than LangChain?
For production agents needing simplicity, local tracing, and Python-native flows, yes. LangChain suits massive integrations; selectools keeps you fast and in control.
