Local by default
Your corpus stays on your machine as Markdown and one SQLite file. Search works offline, and the data shape stays portable.
01
No daemon required
02
No hosted index
03
No data lock-in
Real engine · public sample
Search the docs.
Keep the context.
Mirror documentation to Markdown. Index it with SQLite FTS5. Give people and agents precise, cited answers without a hosted search dependency.
Pure GoLocal-firstApache-2.0Zero AI in this demo
docs-puller / searchfts5
$docs-puller search
"fts5 external content tables" --json
{
"mode": "fts5",
"elapsed_ms": 1,
"results": [
"title": "SQLite FTS5 Extension",
"source": "sqlite",
"path": "fts5.md"
]
}
SQLite · Go · PostgreSQL24 docs ready
One complete path
Docs in.
Context out.
The engine owns the full retrieval loop, so its quality can be tested instead of assumed.
- 01
Pull
Mirror public documentation with stable URLs and manifests.
- 02
Normalize
Turn provider pages into clean Markdown for humans and agents.
- 03
Index
Build a portable SQLite FTS5 index with no external service.
- 04
Search
Return cited snippets through the CLI, local API, or editor.
- 05
Evaluate
Measure Hit@k, MRR, latency, drift, and regressions.
Built for the real work
Retrieval you can inspect, replay, and trust.
Measured
Quality gates included
Frozen fixtures turn retrieval changes into visible Hit@k, MRR, latency, and drift decisions.
Read the methodAgent-ready output
Stable JSON, cited snippets, source filters, and bounded results keep prompts compact and useful.
The public demo has a hard boundary
It uses 24 reviewed public pages. It sends no cookie, account, browser token, or local path. Aggregate runtime counts never contain your raw query.
No install needed
Put the real engine under your own question.
Search SQLite, Go, and PostgreSQL docs in the browser. Then reproduce the same query with the open CLI.