HeM is an open-source governance framework for AI-assisted engineering. It gives AI agents persistent memory, consistent behavior, tiered change management, cross-engine quality gates, and an audit trail — layered on top of whatever methodology you already use.
git clone https://github.com/synthaer/hem.git
cd hem
make installSee QUICKSTART.md to go from install to first governed commit in 10 minutes.
| Problem | Without HeM | With HeM |
|---|---|---|
| Context loss | Re-explain your architecture every session | Resonance loads context automatically |
| Quality variance | Depends on who's prompting | Profiles enforce consistent behavior per project |
| No audit trail | Can't trace which model produced what | Change Management Manifests document every decision |
| Knowledge silos | Trapped in individual sessions | Knowledge promotes across teams and sessions |
| No AI code review | Humans can't review at agent speed | Cross-engine QA Gate — a different engine reviews every PR |
Profiles — Behavior templates that make AI agents consistent across sessions, engineers, and engines. Pick a profile that matches your project type (SaaS, hardware, scripts, open-source, IT ops) and every agent knows the rules.
Tiered Change Manifests — Every commit documents why, not just what. Three tiers: a one-line summary for trivial changes, decisions + self-assessment for feature work, and a full manifest for breaking changes. Agents produce these automatically.
Cross-Engine QA Gate — Before any PR merges, a different AI engine reviews the work. Pattern analysis, defect detection, stub auditing, security scanning, and test coverage — from a perspective the original author doesn't have.
Pre-Commit Governance Hooks — Six automated checks on every commit: change manifest presence, task linkage, credential leak detection, language policy, branch isolation, and scope compliance. No manual steps.
Model-Agnostic — Works with Claude, Codex, Gemini, Copilot, Ollama, Aider, OpenCode — and any future engine. Governance runs at the git layer.
Team Orchestration — Coordinate teams of AI agents across multiple engines. Start with one agent, scale to dozens. The governance adapts.
HeM is not replacing Scrum, Kanban, or SAFe. It's the AI governance layer they don't have.
| Your Methodology | What HeM Adds |
|---|---|
| Scrum | Persistent AI memory, quality hooks, estimation calibration |
| Kanban | Quality gates, cross-engine review, agent governance |
| SAFe | AI agent governance at enterprise scale |
| None | A complete governance framework from solo to team |
One repository, one operating model, every tool can read it:
| Participant | Reads |
|---|---|
| Claude Code | CLAUDE.md |
| Codex / OpenCode | AGENTS.md |
| Gemini CLI | .gemini/AGENTS.md |
| ChatGPT / Copilot / Aider | AGENTS.md or tool-specific config |
| Local models (Ollama, LM Studio) | system-prompts/ + HeM bridge |
| Human team members | COMPACT.md, TASKS.md, project docs |
| Document | What It Covers |
|---|---|
| Quickstart | Install → first commit in 10 minutes |
| Glossary | Complete vocabulary reference |
| Theory of Operation | How HeM works and why |
| Change Management Manifest | Tiered CMD specification |
| QA Gate | Cross-engine review process |
| Best Practices | Practical tips for daily use |
| Philosophy | The heuristic approach to governance |
HeM is an open-source project by Synthaer. The methodology, profiles, hooks, and templates are all MIT licensed.
Synthaer also builds cloud products on top of HeM for teams that want managed infrastructure: hosted persistent memory, fleet orchestration, and quality analytics. HeM works without any cloud product. The cloud products make HeM seamless at scale.
- Method: method.synthaer.ai
- Email: hem@synthaer.ai
- GitHub: github.com/synthaer/hem
MIT License. Copyright (c) 2026 Synthaer.