The verification layer for AI coding agents

Give your agent the right context — up to ~97% fewer tokens

SigMap turns a whole repository into verified function & class signatures — the map an AI coding agent actually needs. No more pasting files or blowing the context window. Try it on any public repo below.

NewMeasured across 405 repos: ~99% fewer tokens, 96× cheaper context — reproducible numbers, with the honest agent results too.See the proof →
97%
fewer tokens
75.6%
hit@5 retrieval
52.2%
more tasks solved
0/21
context overflows (was 16/21)

SigMap v7.25 benchmark · 21 repos · 90 retrieval tasks · sigmap.io

1

Paste a repo

Any public GitHub repository. SigMap detects the real source folders for you.

2

SigMap extracts signatures

Function & class signatures only — secrets redacted, ranked by relevance, full coverage.

3

Feed your agent

Up to ~97% fewer tokens. Copy the context file, ask the codebase, or run npx sigmap in your repo.

Everything SigMap does

full docs ↗

Signature map

try above
npx sigmap

Turns a whole repo into verified function & class signatures — ~97% fewer tokens.

Ranked retrieval

try above
sigmap ask

Ask in plain English, get the exact files that matter — TF-IDF ranked, no LLM needed.

Coverage validation

try above
sigmap validate

Confirms the right files are in scope before you trust the context.

Groundedness judge

in CLI
sigmap judge

Scores whether an AI's answer is actually grounded in your code — the verification layer.

Local learning

in CLI
sigmap learn

Reinforces the files that proved helpful, so retrieval gets sharper over time.

Secret redaction

try above
built-in

Scans signatures for AWS keys, tokens, DB strings and redacts them automatically.

33+ languages

try above
auto

TypeScript, Python, Go, Rust, Java, Kotlin, Ruby, PHP, Swift, C#, C++, Vue, Svelte…

7 agent adapters

try above
--adapter

Emits for Claude Code, Cursor, Windsurf, Copilot, Codex, OpenCode and Gemini CLI.

MCP server

in CLI
sigmap --mcp

Model Context Protocol ready — agents pull ranked context on demand.