The Pipeline
The 10-stage canonical context pipeline.
Every context payload produced by Fuzit passes through the exact same 10-stage deterministic pipeline.
- Discover: Identify repository roots, workspaces, packages, manifests, configuration, Git state, ignored paths, and security policy.
- Scan: Stream files and metadata through repository confinement, ignore rules, sensitive-path policy, and content hashing.
- Index: Store canonical, content-free file records and versioned identities in an atomic incremental index.
- Analyze: Extract normalized symbols, imports, exports, modules, packages, tests, routes, schemas, configuration, and evidence without executing repository code.
- Build Graph: Construct typed nodes and edges for repositories, packages, files, symbols, tests, endpoints, schemas, configuration, docs, snapshots, and changes.
- Enrich: Add framework intelligence, Git evidence, lifecycle facts, profile weights, and detector provenance.
- Retrieve: Score and expand task-relevant context using exact relevance, graph distance, and budgets.
- Explain: Preserve score components, graph paths, profile weights, expansion reasons, budget decisions, index state, omissions, and security decisions. See Explainability.
- Package: Produce deterministic context in supported structured and human-readable formats.
- Serve: Expose the services through CLI, watcher, MCP, VS Code, and plugins.