agno-docs-pp-cli

Offline, agent-native CLI for the Agno developer documentation. One HTTP GET → 3,200 pages indexed in SQLite + FTS5 → ~10-30 ms per query.

agno-docs-pp-cli which "how do teams work"
agno-docs-pp-cli context teams
agno-docs-pp-cli examples "PostgresDb" --language python

---

Install — one command

curl -sSf https://raw.githubusercontent.com/sekai1710/agno-docs-pp-cli/main/install.sh | bash

That script:

1. checks for the Go toolchain (≥1.26)
2. runs go install -tags sqlite_fts5 github.com/sekai1710/agno-docs-pp-cli/cmd/agno-docs-pp-cli@latest
3. runs the first sync (~5 seconds, downloads docs.agno.com/llms-full.txt once)
4. tells you how to use it

If you prefer manual install:

go install -tags sqlite_fts5 github.com/sekai1710/agno-docs-pp-cli/cmd/agno-docs-pp-cli@latest
agno-docs-pp-cli sync

The sqlite_fts5 tag is required — go-sqlite3 ships FTS5 behind it.

Quick tour

agno-docs-pp-cli doctor                              # health check
agno-docs-pp-cli sections                            # map of the docs
agno-docs-pp-cli which "knowledge embedder" -n 3     # top-3 pages
agno-docs-pp-cli context agents                      # full page body
agno-docs-pp-cli examples "team coordinate" --language python

Every leaf command supports --json for agent consumption and --db <path> to override the DB location (default ~/.local/share/agno-docs-pp-cli/data.db).

Commands

Command	What it does
sync	Fetch docs.agno.com/llms-full.txt and rebuild the index. Re-run weekly.
which	Top-N pages for a topic, with snippet. Start here in any agent workflow.
find	Same FTS as which, longer list, CLI-flavoured output.
context	Full markdown body of one page (by slug or full URL).
examples	Paste-ready code blocks. Filter with --language python|bash|json|yaml.
sections	List sections (e.g. models/providers/native) with page counts.
doctor	Pages, examples, last sync time, db path.
version	Print the CLI version.

Agent integration

The CLI is framework-agnostic. --json output drops into anything that can shell out.

Agno (Python) — ready-to-paste

A complete Agno Toolkit lives at examples/agent-toolkit.py. Copy it into your project and:

from agent_toolkit import AgnoDocsTool

agent = Agent(
    model=OpenRouter(id="google/gemini-2.5-flash"),
    tools=[AgnoDocsTool()],
    instructions=["Always call agno_docs_which before answering Agno questions."],
)

Four tools are registered: agno_docs_which, agno_docs_context, agno_docs_examples, agno_docs_sections. All async, all asyncio.to_thread-wrapped, all subprocess-isolated (a CLI crash cannot take down your agent).

Other frameworks

• LangChain / LangGraph — wrap as a Tool via subprocess.run.
• Claude Code / Cursor — call directly from Bash/MultiEdit.
• CrewAI / AutoGen — same subprocess pattern as Agno.
• Shell — agno-docs-pp-cli which … --json | jq.

How it works

docs.agno.com publishes llms-full.txt — every page of the site concatenated into one plain-text markdown file, formatted per the llms.txt convention:

# Approvals
Source: https://docs.agno.com/agent-os/approvals/overview

Manage approval workflows for agents and teams via the AgentOS Control Panel.
...

The CLI is a small pipeline around that single source:

fetch llms-full.txt (10 MB, 1 HTTP GET)
  ↓
parse on "# Title\nSource: <url>" boundaries (one regexp, no HTML)
  ↓
SQLite + FTS5 (porter unicode61, content-table + auto-sync triggers)
  ↓
which / find / context / examples / sections / doctor

Total Go source: ~1,100 lines. No HTML scraper, no headless browser, no auth, no rate limiting — because the source is a single file maintained by the docs team specifically for this purpose.

Why this exists

If an LLM coding agent calls Agno APIs, it needs grounded reference. The three bad alternatives:

1. Let it guess. Hallucinated imports, made-up args, dead URLs.
2. Web-fetch docs.agno.com. Slow, HTML parsing, breaks when the site rebuilds.
3. Scrape and re-scrape. Brittle. Every page change breaks something.

llms-full.txt solves the source problem. This CLI solves the integration problem: turn the 9.9 MB blob into a 30 ms FTS5 query with JSON output suitable for any agent framework.

Comparison with BLZ

BLZ is an excellent general-purpose MCP server for any llms.txt source. If you already use BLZ inside an MCP-aware client (Claude Desktop, Cursor's MCP wiring, etc.), point BLZ at https://docs.agno.com/llms-full.txt and you're done.

This CLI is the complementary path:

Property	BLZ (MCP server)	agno-docs-pp-cli
Transport	MCP stdio	Standalone CLI binary
Setup	Install BLZ + add a source	go install + sync
Sources	Any llms.txt	Agno (this repo) / sibling per-source repos
Integrates with	MCP-aware clients only	Any subprocess-capable runtime
Where it runs	Beside your MCP client	Anywhere you can exec a binary
Domain-specific commands	Generic search	which / context / examples
Typical latency	~50-80 ms	~10-30 ms

Pick BLZ when you want one tool that handles many llms.txt sources in MCP clients. Pick this CLI when you want a single self-contained binary on a server, in CI, or wired into a non-MCP agent.

You can run both side by side.

How it was built

Generated with Printing Press, with one twist on the usual recipe: instead of mirroring a REST API, the generator is pointed at docs.agno.com/llms-full.txt and emits a docs lookup CLI. The llms-full.txt source path is the same insight the BLZ project popularized for MCP search — here applied to a single Go binary.

The repository also preserves the original benchmark methodology under benchmark/ for reference.

Development

git clone https://github.com/sekai1710/agno-docs-pp-cli
cd agno-docs-pp-cli
make build          # ./agno-docs-pp-cli
./agno-docs-pp-cli sync --file ./testdata/llms-full.txt
./agno-docs-pp-cli doctor --json | jq
make install        # → $GOBIN/agno-docs-pp-cli

Or skip make and use go install -tags sqlite_fts5 ./cmd/agno-docs-pp-cli.

License

MIT. See LICENSE.