clay-good/OpenLore

What it solves

AI coding agents are amnesiac and ungoverned. Every task restarts by re-reading the same files, long sessions drift onto stale assumptions, and nothing warns the agent when a change is about to break a contract or cross a boundary. OpenLore fixes both halves: it gives agents persistent, code-anchored memory so they start every task already oriented, and it adds guardrails that certify changes are safe before they land — all without an LLM in the loop.

How it works

A one-time static analysis builds a live knowledge graph of your repo — call structure, types, tests, decisions, and infrastructure-as-code. After that, the agent gets oriented automatically at the start of each session via MCP tools like orient(), and governance tools such as certify_public_surface, change_impact_certificate, and verify_claim check diffs for breaking changes, sensitive boundaries, and groundless claims. Everything runs locally, deterministically, with no API key. A file watcher re-indexes as you edit, and any fact that can't be recomputed in time is explicitly marked stale rather than silently wrong.

Who it’s for

It's a strong fit for codebases too large to hold in your head (and the model's), private or niche code the model never memorized, long sessions where stale assumptions compound, and polyglot repos or code plus the IaC that deploys it. It's worth trying and measuring on mid-size repos, and it's probably not the right tool for small, familiar repos answering shallow questions — the agent's built-in search is cheaper there, and OpenLore publishes the measurement that says so.

Highlights

• Deterministic and fully local: no API key, no network calls, same answer every time.
• One-time openlore install auto-detects your agent, wires the MCP server, and indexes the repo — even new repos you open later.
• Memory: orient() returns relevant functions, callers, specs, tests, and insertion points in ~430µs p50 on a 15k-node graph.
• Governance: breaking-change verdicts with named consumers, sensitive-boundary checks, grounded claim verification, and a commit gate — all advisory by default with no API key.
• Benchmarks show 7–21% cost savings and 26% fewer round-trips on deep tasks in large repos; losses are published honestly alongside wins.
• Supports 21 programming languages and 12 Infrastructure-as-Code ecosystems on one shared graph.