Code Intelligence · Mind RAG

Search by meaning, not by string match.

Semantic code search. Returns citations with file position and confidence scores. We use it ourselves — to navigate KB Labs and give AI agents context without hallucinations.

Documentation Install KB Labs

Indexing

Incremental by content-hash.

Terminal

$ kb mind index

  ✓  Index updated in 22s (incremental)

  Files      discovered: 1 840 · processed: 68 · skipped: 1 772
  Chunks     stored: 9 440 · updated: 480 · rate: 7.06/file
  Cleanup    deleted files: 3 · deleted chunks: 22

Source filesAll .ts, .tsx, .js, .jsx files in tracked packages

Index backendQdrant vector store (local or remote)

Embedding modelConfigured via adapter — openai, cohere, or custom

Re-index triggerAutomatic on kb-dev start or manual via pnpm kb mind reindex

Search modeSemantic (dense), keyword (sparse), or hybrid

Source verification

Not in the index — it'll say so honestly.

Every answer is tied to real chunks from the index with exact file and line numbers. The system cannot cite a source that isn't in the codebase.

If nothing is found — returns completeness: "minimal" and an explicit message. If found but uncertain — LOW_CONFIDENCE warning with a percentage. Low confidence results are not cached.

json

$ kb mind search --text "как работает pipeline релиза" --agent // i18n-ignore

{
  "chunks": [
    {
      "path": "plugins/release/manager-core/src/pipeline.ts",
      "span": { "startLine": 1, "endLine": 12 },
      "score": 0.94,
      "text": "Flow: plan → snapshot → checks → build → verify..."
    },
    {
      "path": "plugins/release/manager-cli/src/cli/commands/run.ts",
      "span": { "startLine": 180, "endLine": 210 },
      "score": 0.87,
      "text": "const result = await runReleasePipeline({..."
    }
  ],
  "contextText": "Release pipeline состоит из 8 последовательных шагов...", // i18n-ignore
  "meta": { "schemaVersion": "agent-response-v1", "timingMs": 14200 }
}

Commands

For humans and agents.

pnpm kb mind search --text "how does X work"Semantic search across the full codebase.

pnpm kb mind search --text "RoutingBackend" --agentAgent-mode output: JSON lines with sources and confidence.

pnpm kb mind reindexForce full re-index of the codebase into the vector store.

pnpm kb mind reindex --affectedRe-index only files changed since last index run.

pnpm kb mind statusShow index health: total chunks, last indexed, backend status.

pnpm kb mind configShow active embedding model and vector store connection.

AI agents

First-class tool for agent workflows.

--agent returns structured JSON with schemaVersion: "agent-response-v1". Agents get exact citations with file position — no guessing context.

Three modes: instant — fast, auto — balanced, thinking — reasoning chain before the answer.

InputNatural language query or structured JSON prompt

RetrievalTop-K semantic search with confidence threshold

Context windowSources injected into LLM context automatically

OutputStructured answer with file citations

AuditEvery search logged with query, sources, and latency

ModelConfigured via Gateway — swap without code changes

Install

Semantic search. Honest answers.

kb marketplace install @kb-labs/mind-entry

Documentation Install KB Labs