← Back to Discover

Mneme HQ

Open-source Python tool that records architectural decisions and enforces them as deterministic checks on AI coding agents, in repositories and in CI.

Code reviewCommand-line toolsMCP servers
Visit website ↗

What it does

Mneme keeps a structured, auditable record of engineering decisions and uses it to guide and check what AI coding agents propose. Its first supported decision domain is architecture: it turns architectural decisions and ADRs into guardrails across coding agents, repository changes, generated rules and CI gates, so decisions stay active instead of sitting as passive documents. Enforcement is deterministic. The same input and decision state give the same allow, warn or fail result, and the core verdicts do not rely on an LLM judge. Mneme applies checks at the earliest boundary each workflow exposes: before generation, before supported file edits, after edits through a working-tree audit, and before merge through a CI gate. An Architecture Audit reports which decisions are already protected, ready to protect, in need of modelling, or guidance only. Mneme requires Python 3.11 and installs with pip. The mneme command initialises a project's decision memory, records decisions, checks a proposed input in strict mode with a matching exit code, and has a setup mode that enables no blocking enforcement. A local MCP server exposes six tools for proposing and querying decisions without the power to accept them. Documented integrations include Claude Code, the Claude Agent SDK, Google Antigravity, Codex CLI and Kiro CLI, with rules export for Cursor, CI gates for GitHub Actions and GitLab CI, and an experimental OpenCode integration. The code is released under the MIT license.

Capabilities

  • Structured record of architectural decisions built from ADRs
  • Deterministic allow, warn or fail checks without an LLM judge
  • Architecture Audit showing which decisions are actually protected
  • Integrations for Claude Code, Codex CLI, Kiro CLI and others, plus CI gates
  • Local MCP server with six tools for proposing and querying decisions
  • Setup mode that adds no blocking enforcement

When to use it

  • Your team uses AI coding agents and wants architectural decisions enforced rather than only documented
  • You want a repeatable CI check for architectural rules
  • You need an audit of which ADRs are enforced mechanically today