Short answer: Brigade ships a built-in evidence ledger, formerly the standalone MiseLedger: the local evidence archive, with Brigade’s evidence commands to plan crawls and export verify/run receipts into it. Crawl and import stay operator-run at a process boundary; Brigade does not silently upload logs.
The problem
Agent work leaves scraps: session transcripts, verify receipts, git commits, chat exports. Grepping folders does not scale. Hosted “agent analytics” products require trust and network access you may not want for private work.
The path
brigade setup # installs the evidence engine, digest-verified
brigade evidence crawl plan # review-only: see the init/crawl/doctor commands first
brigade evidence crawl sessions # crawl harness session logs into the ledger
brigade evidence search "auth timeout"
brigade receipts export miseledger --target . --new-only --import
brigade evidence doctor
Pipeline in plain language: crawl local sources → adapter JSONL → the Evidence SQLite ledger + FTS → evidence briefs on the next Brigade run.
What stays a boundary
- The Evidence engine is built into Brigade and installed by
brigade setup, but it runs at a process boundary: a Go binary, not code inside the Python package. brigade evidence crawlandsearchexecute the engine only when you invoke them. Theplancommands stay review-only.- Receipt export is local and fail-open when the engine is missing.
When this is enough
Use Brigade’s evidence ledger when you want local FTS over agent history and receipt-linked evidence on future runs. Use a hosted observability product when you need multi-tenant dashboards, SSO, or team-wide cloud retention policies.
Questions
Is this the same as memory handoffs? No. Handoffs are durable notes you promote into memory cards. The evidence ledger is searchable history and receipts, treated as untrusted context rather than instructions.
Does it work with the code graph? Yes. Code intelligence covers structural code deltas; Evidence covers session and receipt evidence. Together they close the receipts-to-context loop.
Where is the capability page? brigade.tools/evidence-memory. CLI: brigade evidence.
Related: Code intelligence / code graph, MCP sync.