Quick start

This page is the fastest path to the real v5.5 workflow.

1. Install or run directly

npx sigmap

You can also install globally:

npm install -g sigmap
sigmap

2. Generate the base context

sigmap

That writes the default context file and builds the signature index your tools will use.

3. Ask a real question

sigmap ask "explain the auth flow"

This creates a focused context file for the current task instead of forcing the model to reason over the whole repo.

4. Validate the coverage

sigmap validate --query "auth login token"

If coverage is low, fix that before trusting the answer.

5. Get the AI response

Copy the context and ask your AI assistant:

# Display the focused context (already created in step 3)
cat .context/query-context.md
# → Copy this into Claude, Copilot, ChatGPT, or your IDE chat
# → Ask: "Explain the auth flow" (paste the context first)
# → Copy the AI's answer

Save the AI's response to a file:

# Create response.txt with the AI's answer
cat > response.txt << 'EOF'
<Paste the AI's answer here>
EOF

6. Judge the answer

Score whether the AI's answer is grounded in your code:

sigmap judge --response response.txt --context .context/query-context.md
# Output: Score (0.0–1.0) and PASS/FAIL indication

That tells you whether the answer looks supported by the supplied code context.

7. Optional: Stay fresh with automation

If you want SigMap to stay fresh in the background:

sigmap --setup
sigmap --watch

--setup installs the git hook and MCP config. --watch is best during active coding.

8. Optional: Local learning

If a file was especially helpful or misleading, reinforce that locally:

sigmap learn --good src/auth/service.js
sigmap weights

Next steps

• Daily workflow: ask, validate, judge, learning
• Reference: CLI, Config
• Integrations:
  • Open-source agents — OpenCode, Aider, OpenHands, Cline
  • Local LLMs — Ollama, llama.cpp, vLLM (no API keys)
  • MCP server — Claude Code, Cursor, Windsurf
  • Repomix integration
• Proof: Benchmark overview