Tools for your agent.
An assistant of your own. A whole team of them.
this one ~10 MB Rust binary. It gives your agent real context — search the web, read any page, and curate your own knowledge — then ships Big Smooth, an always-on AI assistant on your machine, and the coordination to run a whole team of agents at once.
$ brew install SmooAI/tools/th
$ th upGive your agent context: the web, a page, your knowledge.
Three commands, one job — feed an agent real context. Search the open web, read a specific page as clean markdown, or run the very same semantic retrieval your agents use over your org's own knowledge base. Each is one th command, usable by an AI coding agent or by you.
Search the web
th search "…"Ranked results with an optional synthesized, cited answer — served by our own search stack, no third-party vendor.
Read a page
th crawl scrape <url>Any URL → clean markdown, via a real browser that gets pages a plain fetch 403s on.
Query your knowledge
th knowledge search "…"The SAME semantic retrieval your agents run over your org's own documents — from the CLI. Feed it with th knowledge add-url.
th search ships with an anonymous free tier and full depth once you sign in; high-volume th crawl unlocks with a paid plan.
Preview — web crawl is not yet generally available; the limits and prices above are proposed, not final.
Big Smooth ships the diff — on your machine.
Not a chatbot that hands you snippets. A chat-first, always-on agent that reads, writes, and runs — in a streaming terminal (th code) or your browser (th web) — on the smooth-operator engine, on the models you choose, watched by in-process safety the whole way.
It does real work
Read, search, write, edit, run shell, dispatch coding tasks and drive them to green.
On your machine
One binary, foreground or background service. Your files, your logins — not someone else’s cloud.
Safe by default
Narc judges every tool call — regex fast-paths plus an LLM judge — and a permission engine gates the rest with allow / ask / deny.
Same agent — on your laptop or at fleet scale.
The engine Big Smooth runs on your machine also runs hosted in the cloud: org agents served over the platform API, fronted by a gateway that reaches every major model. Start local, then scale to a hosted, always-on fleet — same agent, same tools, no rewrite.
On your machine
th up · th code · th web — your files, your keys, zero setup. Great for hands-on work and tight loops.
Hosted in the cloud
th api agents serves the very same engine to your whole org — scaled, always on, and reachable over the platform API.
Install new powers for your agent.
Search a marketplace of extensions — the index that ships with th plus npm packages tagged for Smooth — and install one with a single command. Big Smooth loads installed extensions per turn, so a new tool is live on the very next message. No rebuild.
$ th ext search browser
$ th ext install npm:<package> # or a local dir, or git:user/repoOne agent is good. A coordinated team ships faster.
Point several AI coding agents at the same codebase — in any harness, on any machine — and th is the fabric that keeps them in sync: agents mail each other, pull from one shared work queue, and hand off cleanly instead of stepping on each other.
Agent mail
th agent register names a session; th msg send --to <name> passes work, context, and “it’s green” between agents — harness-agnostic, any machine.
One shared work tracker
th pearls is a dependency-aware queue backed by embedded Dolt. Every agent pulls ready work, closes what it finishes, and syncs over git — with two-way Jira sync.
Drop into your agents
A plugin brings the th workflow — mail, pearls, org skills, repo guardrails — straight into popular AI coding agents. Nothing to hand-copy.
$ th agent register --name reviewer
$ th msg send --to builder --body "tests green — ship it"
$ th pearls ready # shared, dependency-aware queueThe whole platform, one command away.
Reach for thbefore curl, before the dashboard. Config & secrets, LLM-gateway keys, auth clients, orgs, agents, knowledge, observability, testing, and work-tracking — all behind one binary, each command self-documenting.
Every model. One key. One bill.
89 models from every major lab — frontier and fast, open and closed — behind one OpenAI-compatible endpoint. Route each turn to the model that fits and pay per token across every provider. Skip the flat per-seat subscription.
Cheap by default
Send heavy work to fast open models at a fraction of frontier cost; reach for a frontier model only when it earns it.
No per-seat lock-in
One org key, billed by token. No monthly minimum per developer.
Switch models, not code
It speaks the OpenAI API — change one string to move providers.
Open source. On your machine. Zero runtime deps.
One ~10 MB Rust binary, MIT-licensed. No Docker, no Node runtime, no per-project setup ritual. It runs where you run it — macOS and Linux today, Windows via WSL while native support is in flight.
One binary
~10 MB, zero runtime dependencies. macOS and Linux.
Your infrastructure
Foreground process or background service — your files, your keys.
Built in the open
MIT-licensed on GitHub. Read it, fork it, extend it.
Install once. Run the platform.
One 10 MB binary. th search, th config, and th apiare ready the moment it's installed; th up starts Big Smooth.
$ brew install SmooAI/tools/th
$ th up