- Typed core contracts
- Adaptive request router
- Orchestration layer
- Context budget enforcement
- Local project memory
- Repository intelligence
- Multi-language symbol intelligence
- Validation engine
- CLI and diagnostics
- Integration tool bridge
- Local dashboard
- Incremental repository index
- Dependency and call graph
- Impact analysis
- Git-aware change intelligence
- Automatic project knowledge extraction
- Query-aware context ranking and deduplication
- Query result caching
- Native MCP stdio transport
- One-command project setup
- Reproducible benchmark suite
- CI across supported Python versions
- Deeper language-aware parsing and type resolution
- Cross-file call resolution with ambiguity scoring
- Incremental graph updates without full graph rebuilds
- Semantic retrieval providers
- Git-aware symbol history and blame intelligence
- Architecture pattern inference with confidence scores
- Persistent benchmark history and regression gates
- Multi-repository context
- Shared local team memory
- Pull request intelligence
- Architecture drift detection
- Agent task traces
- Reproducible external evaluation suites
- Guarded language-server semantic editing
- Live coding-agent MCP E2E harness
- Native Tree-sitter parser providers with structural fallback
- Warm persistent backend sessions and concurrent orchestration
- Live backend conformance against pinned revisions
- Docker Core/Full integration CI
- Expanded real-repository benchmark corpus and CI artifacts
- Signed/attested release pipeline and automated package/container publishing
- Security audit, CodeQL, dependency review, SBOM, and path-boundary tests
- Observatory dashboard for traces, drift, graph health, benchmark history, and PR risk
- Canonical package/brand identity policy
- Remote shared-memory synchronization with conflict resolution
- Persistent vector-database providers for very large repositories
- Hosted/remote MCP transport with authentication, TLS, quotas, and access policy
- Multi-node indexing and retrieval workers
- Published longitudinal performance history from scheduled reproducible runs
- Organization-level policy, audit retention, and workspace administration
- Unified evidence model with categorical trust tiers and provenance
- Central ranking, deduplication, and conflict policy
- Precision code intelligence: exact definitions, references, implementations
- Precision graph fusion with staleness detection and conservative fallback
- Dependency inventory across Python, Node, Rust, Go, JVM, and .NET manifests
- Declared-constraint and resolved-version separation
- Optional, disabled-by-default version-aware documentation provider
- Structural search with typed matches and captures
- Guarded structural rewrite: preview, content-hash transaction, rollback, validation
- Evidence-aware impact analysis and context pipeline
- Router evidence planning that keeps local questions local
- Measured heuristic-versus-evidence benchmark cases
A public release is considered evidence-backed only when the exact release commit has passed Core CI, backend conformance, parser-provider tests, Docker integration, security checks, and the relevant benchmark run. Credential-gated external services must be reported as skipped rather than silently treated as passed.