Search BrigadeStack

Search BrigadeStack pages and docs.

BrigadeStack
Menu

Comparisons

Local code graph for AI coding agents

You want agents to answer callers, callees, and impact without shipping your repo to a hosted code-intel service.

Updated November 15, 2026

Short answer: BrigadeStack ships a built-in code graph (see Code Intelligence): a local SQLite code graph over the CLI and an optional MCP server. BrigadeStack attaches code-graph deltas to verify and run receipts when the graph is present. The graph is optional.

The problem

Agents reason about code from whatever fits in context. Grep finds strings. It does not answer “who calls this?” or “what broke when this symbol changed?” Hosted code-intel tools help, but they move your tree off-machine and do not plug into a local receipt loop.

The path

brigadestack setup             # installs the code-graph engine, digest-verified
brigadestack code sync         # builds .brigade/code-graph/code-graph.db in the repo
brigadestack code context "auth middleware" --markdown
brigadestack code impact <symbol>        # blast radius of changing one symbol
brigadestack code affected               # tests statically tied to your changed files
brigadestack search doctor     # advisory code-graph + optional code-search health
brigadestack work verify run --target . --command "pytest -q" --capture brigade-work

The code graph owns the structure. BrigadeStack owns the operator loop: when a graph is present, verify and run receipts can carry a compact code_graph_delta so the next session sees what actually changed. In the 0.31 line, Claude Code work-loop hooks also track the files an agent has read and record callers it has not seen before an Edit or Write. The check warns and does not block the edit, and a missing or incomplete graph records a null result with a reason.

BrigadeStack surface

The code graph ships with brigadestack setup, and brigadestack code (sync, callers, callees, impact, context, outline, affected, export, diff) is the query surface. Setup installs the engine as brigade-code-graph, and new repos index into .brigade/code-graph/. Older engine names and an existing old index still work, with deprecation notices, until 0.33.0. A compatibility path also wires it through the search component:

  • brigadestack add search (a compatibility path after brigadestack setup)
  • brigadestack search status / doctor / sync plan (plans only, and it does not run sync for you)

Optional semantic search (code-search-api / MCP) sits beside the graph and does not replace it.

When this is enough

Use BrigadeStack’s code graph when you want local callers/callees/impact and structural proof on verify receipts. Use a full language server or hosted code-intel product when you need deep IDE features, multi-language enterprise indexes, or team-wide cloud search.

Questions

Does BrigadeStack require a code graph? No. The code graph is optional and fail-open. Absent tools never break a workspace doctor run.

Is the graph uploaded anywhere? No. The database lives under .brigade/code-graph/ in the target repo. An index built by an older release stays where it is and is used in place.

Can agents query it over MCP? Yes. The code graph ships a read-only MCP surface, and BrigadeStack can also sync MCP catalogs with brigadestack mcp sync.

Capability page: Code Intelligence on brigadestack.com. CLI: command docs.