How jayasuriyasr/codeatlas works
SEI is a repository-level GraphRAG investigation engine for Python/Django and TypeScript/Next.js codebases. Its main workflow indexes source into a Neo4j symbol graph and embeddings, routes developer questions to structural, semantic, or trace retrieval, packs relevant code into model context, then streams a cited answer with grounding checks. The graph reflects documented capabilities and the sampled query/indexing seams; several internals and integrations are represented only at directory level because their implementations were not excerpted.
Main entry points: indexer.py Index coordinator · app.py Investigation API · cli.py CLI wiring · router.py Query router
Components 25
Source indexing
- Index coordinator
index/indexer.py - Python parser
adapters/python.py - TypeScript parser
adapters/typescript.py - Symbol model
adapters/base.py - Chunk and scrub
index/chunker.py - Embedding cache
index/cache.py - Embedding providers
index/providers.py - Move detection
index/moves.py
Graph and retrieval
- Graph writer
graph/writer.py - Graph reader
graph/reader.py - Neo4j
Answer generation
- Query router
retrieve/router.py - Hybrid retrieval
retrieve - Frame resolution
retrieve/frames.py - Context packer
pack/packer.py - Code renderer
pack/render.py - Answer streaming
api/stream.py - Grounding checks
ground/validate.py - Claim classifier
ground/classify.py
API and evaluation
- Investigation API
api/app.py - CLI wiring
api/cli.py - Runtime metrics
metrics.py - Evaluation harness
eval/harness.py
Other components
- Developer
- Code repository
Connections 27
- Developer → Investigation API: asks question
api/app.py - Investigation API → Query router: routes query
api/app.py - Investigation API → Context packer: packs context
api/app.py - Investigation API → Grounding checks: validates answer
api/app.py - Investigation API → Answer streaming: streams events
api/app.py - CLI wiring → Index coordinator: configures indexing
api/cli.py - CLI wiring → Investigation API: configures service
api/cli.py - Code repository → Index coordinator: provides source
README.md - Index coordinator → Python parser: parses Python
index/indexer.py - Index coordinator → TypeScript parser: supports TypeScript
README.md - Index coordinator → Chunk and scrub: builds chunks
index/indexer.py - Index coordinator → Embedding cache: embeds through
index/indexer.py - Index coordinator → Embedding providers: uses provider
- Index coordinator → Move detection: collects moves
index/indexer.py - Python parser → Symbol model: produces symbols
adapters/python.py - TypeScript parser → Symbol model: produces symbols
README.md - CLI wiring → Graph writer: configures writer
api/cli.py - Graph writer → Neo4j: writes graph
graph/writer.py - CLI wiring → Graph reader: configures reader
api/cli.py - Graph reader → Neo4j: reads graph
- Query router → Hybrid retrieval: selects retrieval
retrieve/router.py - Query router → Frame resolution: resolves trace
retrieve/router.py - Investigation API → Hybrid retrieval: retrieves candidates
api/app.py - Context packer → Code renderer: renders blocks
pack/packer.py - Answer streaming → Claim classifier: classifies claims
api/stream.py - Grounding checks → Claim classifier: uses classification
ground/validate.py - Evaluation harness → Runtime metrics: checks metrics