clay-good/OpenLore

What it solves

AI coding agents often struggle with "amnesia" and a lack of governance. They frequently re-read the same files for every task, rely on stale assumptions during long sessions, and may introduce breaking changes or cross sensitive architectural boundaries without warning. OpenLore provides a deterministic, local-first memory and guardrail system to ensure agents are oriented and their changes are safe.

How it works

OpenLore performs a one-time static analysis of a repository to build a live knowledge graph encompassing call structures, types, tests, and Infrastructure-as-Code (IaC). It operates as an MCP server with no LLM in the hot path, meaning it provides grounded, deterministic answers rather than probabilistic guesses.

Agents use the orient() call to immediately identify relevant functions, callers, and insertion points for a task. For governance, it uses static analysis to certify public surfaces for breaking changes, flag unauthorized paths into sensitive boundaries, and verify claims with citations. It can also be integrated as a commit gate via openlore enforce to block unsafe changes.

Who it’s for

• Developers working with codebases too large for a model's context window or a human's memory.
• Teams using private or niche code that LLMs haven't been trained on.
• Users of AI coding agents (like Claude Code, Cursor, Cline, Continue) who need to prevent agents from introducing architectural regressions.
• Polyglot projects that combine application code with IaC.

Highlights

• Deterministic Memory: Replaces exploratory file reads with a single orient() call that returns relevant code and call paths.
• Static Analysis Guardrails: Certifies breaking changes and flags boundary violations without relying on an LLM.
• Local-First: No API keys required for core functionality; source code never leaves the machine.
• Broad Support: Supports 21 languages and 12 IaC ecosystems.
• Agent Integration: Auto-detects and wires into common AI coding agents and provides an MCP server.