One repository.
Many engineering signals.
Vericore connects repository structure, dependencies, Git evolution, changes, architecture, state, evidence, planning, governance, and verification into one local-first engineering intelligence layer.
Repository Intelligence
Java and Kotlin discovery and parsing with configurable exclusions, file limits, parallel parsing, content-based caching, and self-contained HTML reports.
Dependency & Graph Intelligence
Dependency relationships, wildcard-import handling, connected components, cycle detection, PageRank knowledge hotspots, and learning-path generation.
Git & Evolution
Authorship, churn, modification time, recent changes, deterministic time-based commit sampling, source-history metrics, and evolution hotspots.
Change Impact
Reverse-dependency traversal combines graph depth, PageRank, churn, and package signals to expose affected areas and test candidates.
PR Intelligence
Structured pull-request change context including change size, risk signals, architecture findings, test recommendations, and machine-readable artifacts.
Architecture Intelligence
Architecture relationships, cycles, drift against explicit baselines, deterministic contracts, severity limits, and decision records.
Engineering Context
Versioned snapshots, repository-relative fingerprints, stable digests, Git HEAD capture, dirty-state information, and snapshot diffs.
Engineering Reality
Deterministically composes analysis and repository context so stale analysis can be rejected instead of silently mixed with newer source state.
Grounded Evidence & Q&A
Versioned evidence contracts, bounded evidence generation, citation paths, intent classification, deterministic ranking, and insufficient-evidence handling.
Engineering Planner
Read-only plan synthesis from grounded evidence with affected components, risks, steps, evidence IDs, verification criteria, uncertainties, and contract binding.
Agent Change Contracts
A versioned repository-bound artifact binding canonical repository identity, prepared HEAD, paths, components, evidence, verification, architecture expectations, and SHA-256 fingerprint.
Prepare → Change → Verify
Prepare establishes the persisted safety boundary. Implementation happens independently. Verify checks the original contract and detects tampering, mismatch, repository changes, and stale HEAD.
Provenance & Decision History
Stable provenance IDs, repository HEAD capture, explicit unknown-Git state, prepare/verify provenance, and idempotent architecture decision persistence.
CLI · REST · MCP
Local CLI workflows plus Ktor REST and MCP stdio surfaces with path validation, remote-URL rejection, rate limiting, and sanitized public errors where applicable.
Optional AI Assistance
Configured Gemini/Anthropic providers can reason over bounded repository-derived context. AI is disabled unless configured and is not repository truth.
Deterministic evidence first.
Repository-derived facts form the evidence boundary. Engineering Context and Engineering Reality make state explicit. Grounded Q&A and planning consume bounded evidence, while optional AI adds reasoning without becoming the source of repository truth.
Prepare before you change. Verify what you actually changed.
The persisted Agent Change Contract is the verification boundary. Verification does not silently reconstruct a replacement contract from a mutable plan after preparation.
Know the boundaries.
- Analysis currently focuses on Java and Kotlin.
- Kotlin parsing has known complex-syntax limitations.
- The planner is read-only; autonomous source modification is not implemented.
- REST and MCP are trusted local/internal integrations without authentication or tenant isolation.
- Remote repository URLs are not accepted by local server endpoints.
- Architecture contracts are deterministic governance artifacts, not human approval workflows.