From c8d894497b370514c30c9978ffd77f0086c8127d Mon Sep 17 00:00:00 2001 From: Brent Rager Date: Mon, 13 Jul 2026 15:21:06 -0400 Subject: [PATCH] =?UTF-8?q?docs:=20README=20=E2=80=94=20document=20th=20se?= =?UTF-8?q?arch=20/=20th=20crawl=20/=20th=20knowledge=20agent=20tools?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The canonical CLI reference covered none of the three agentic-context commands. Add an 'Agent tools — search, crawl, knowledge' section: three ways to feed an agent context (the web, a page, your org's knowledge), each with examples and the free/authed tier note. Co-Authored-By: Claude Opus 4.8 (1M context) Claude-Session: https://claude.ai/code/session_01Vg7frHqHSAUXhQiRNdr3NW --- .changeset/readme-agent-tools.md | 5 +++++ README.md | 29 +++++++++++++++++++++++++++++ 2 files changed, 34 insertions(+) create mode 100644 .changeset/readme-agent-tools.md diff --git a/.changeset/readme-agent-tools.md b/.changeset/readme-agent-tools.md new file mode 100644 index 00000000..ccfb0e6a --- /dev/null +++ b/.changeset/readme-agent-tools.md @@ -0,0 +1,5 @@ +--- +"@smooai/smooth": patch +--- + +README: document the agent-tools trio — `th search` (web), `th crawl` (a page), and `th knowledge search` (your org's knowledge base). The CLI reference previously covered none of them. diff --git a/README.md b/README.md index 994b1fb2..4e3e383d 100644 --- a/README.md +++ b/README.md @@ -302,6 +302,35 @@ Global config lives at `~/.smooth/`; project config at `/.smooth/`. Project entries shadow global on name collision. See [`docs/extending.md`](docs/extending.md) for the full guide. +### Agent tools — search, crawl, knowledge + +Three ways to feed an agent real context — the open web, a specific +page, and your org's own knowledge. Each is one command, each usable by +`th code` or by you, each with a bundled free tier (authenticate with +`th auth login` for the full tier). + +```bash +# Search the web → ranked results (+ a synthesized, cited answer). +# Our own search stack (self-hosted SearXNG + in-house crawler + LLM +# synthesis) — no vendor. --answer / --depth advanced / --max are the +# authed tier. +th search "rust async traits" +th search "postgres upsert on conflict" --answer --max 8 + +# Read a single page as clean markdown, through a real browser (JS render +# + browser UA) — gets pages a plain fetch 403s on. +th crawl scrape https://example.com/some/article + +# Search your ORG's knowledge base — the SAME semantic retrieval your +# agents run over your own documents (embed → dense+sparse hybrid → RRF). +# Scope to one document with --doc. (Authed; knowledge is org-private.) +th knowledge search "refund policy" +th knowledge search "pricing" --doc + +# The rest of the knowledge surface (alias: th kb): +th knowledge list | show | content | upload | website | process | delete +``` + ### Run a pearl (`th run`) Dispatch a pearl (or ad-hoc prompt) to a Smooth operative. Big Smooth