Production-ready, open-source infrastructure for AI agents.
The reliability layer between "the agent works on my machine" and "the agent works in production."
Open source · MIT licensed · Self-hostable · No telemetry · No feature-gated "community edition"
AI agents are easy to demo and hard to run in production. They forget everything between sessions, break silently when a prompt changes, ship without anyone knowing if they actually work, burn money no one can account for, and share credentials no one can revoke.
Emart AI builds the open-source stack that closes those gaps — the unglamorous infrastructure that stands between a clever prototype and something a team can actually depend on. Everything is MIT-licensed, self-hostable, and free forever at the core. Commercial offerings, when they come, are additive — never a tax on the parts that matter.
We're building it in the open, from Lagos, as a community — because the people who most need a way into AI engineering are the ones least served by the tools built somewhere else. More on that below.
Teams deploying agents hit the same failures, in the same order:
- Session memory loss — agents start from zero every session.
- Silent prompt regressions — one prompt tweak quietly breaks another task.
- Deployment confidence gaps — no way to know an agent works before users find out.
- Cost & outcome opacity — you can see the bill, not whether it was worth it.
- Insecure credentials — one leaked key, and there's no way to revoke just one agent.
Six focused tools, one coherent stack.
| Project | What it does |
|---|---|
| Remembr | Shared memory layer for multi-agent systems — persistent context across sessions, agents, and time. pip install remembr |
| Mindr | Codebase memory for AI coding agents (Claude Code, Cursor, Aider) — your agent stops forgetting your project. npm install -g mindragent |
| Project | What it does |
|---|---|
| Evalflow | "pytest for LLMs" — catch prompt regressions in CI before they ship. pip install evalflow |
| PlayAgent | Pre-production agent testing — find failures before your users do. |
| Project | What it does |
|---|---|
| Spendline | Cost and outcome tracked together, per agent run — not just what you spent, but what you got. |
| Scopeform | Scoped, individually revocable credentials for agents — one-click revoke, no key rotation for everyone else. |
New here? Start with Mindr — it works offline, needs zero setup, and you'll feel the point in about a minute.
- Open core, forever. The self-hostable stack is fully functional with no account, no telemetry, and no features held hostage behind a paywall.
- Public by default. Every release is public, always. Roadmaps, decisions, and progress happen in the open.
- Real infrastructure. Tested, documented, versioned, and built to be trusted in production — that's the bar, and we hold it.
Emart AI is also a community, and this is the part we care about most.
Most people learning AI engineering are stuck doing tutorial projects nobody will ever look at. We think there's a better way: learn by contributing to real infrastructure that engineers around the world actually use — and walk away with proof you can show.
Developers breaking into AI engineering. Students. Self-taught builders. Anyone who wants to learn how memory layers, evals, agent testing, cost tracking, and MCP tooling actually work — by building them, not just reading about them. You don't need to be an expert. You need to be willing to learn.
- Real, merged contributions to software people use — the kind of work that belongs on a résumé, not a to-do-app tutorial.
- Public recognition, every time. Every merged contribution gets a shoutout across our channels. Your work, with your name on it, in public.
- Mentorship and review from maintainers and from contributors a few steps ahead of you — and the chance to become a reviewer and mentor yourself.
- A learning path, not a maze. Issues are labeled by what they teach you, so contributing doubles as a curriculum in AI engineering.
- Pick a repo that interests you and look for issues tagged
good first issueandlearns:. - Open a PR. Ask questions freely in the issue or in Discussions — there are no dumb ones here.
- Get merged, get your shoutout, and pick the next one.
We're warm to people and strict about code — that's how a place stays genuinely welcoming and builds infrastructure worth trusting. Come learn, ship something real, and grow with us.
Everything happens in the open — discussions, releases, and roadmaps.
- Twitter/X: @emart_ai
- Discord & WhatsApp community: launching soon — follow @emart_ai for the invites.
Built by Emmanuel (@Emart29), a Data Scientist and ML/AI engineer working the infrastructure layer between model training and production. Based in Lagos — building Emart AI as open infrastructure, and a community where more people can learn to build it too.
Everything here is MIT licensed. Use it, fork it, self-host it, build on it.