memrain

<div align="center"> # Memrain **Long-term memory for your AI agents.**<br> Your notes, decisions and code, searchable from Claude, ChatGPT and Codex, with the source attached to every answer. <img src="docs/assets/hero.webp" alt="Notes, a code file, a chat bubble and a sticky note flow into one filing cabinet, which feeds answers to three laptops." width="100%"> [](./LICENSE)   </div> **For developers and small teams who work with AI agents every day.** Memrain is free and open source (MIT). It runs in your own AWS account, so you pay only your AWS bill: about $52 a month for the server and database, plus model usage if you turn on the paid features. **To start, follow [docs/QUICKSTART.md](./docs/QUICKSTART.md).** ## Why - **Your agent forgets.** Every new chat starts from zero, so you paste the same context again and again. - **Your knowledge is scattered.** Decisions sit in notes, the reasons in old chats, the details in code. - **Answers without sources are guesses.** You cannot check an answer that does not say where it came from. Memrain reads all of it, stores it in one place, and lets every agent you use search it. Each result points to the exact page it came from. ## What you get | | | |---|---| | <img src="docs/assets/feature-memory.webp" alt="An answer card linked by a thread to the highlighted line of the note it came from." width="260"> | **Answers you can check.** Ask "what did we decide about the auth flow?" and get the passages from your own notes, each with the page it came from. Your agent writes the answer; Memrain supplies the evidence. | | <img src="docs/assets/feature-code.webp" alt="One highlighted function in a code file, with lines fanning out to the files that call it." width="260"> | **It understands your code.** Who calls this function, what breaks if I change it, where is it defined: for TypeScript, Python and Go. Bash and SQL files are searchable too. | | <img src="docs/assets/feature-team.webp" alt="One filing cabinet with three locked drawers in different colours, each with its own key." width="260"> | **One memory for the team, a private space for each person.** One connector for everybody, a separate source per person, a daily spending cap, and access you can revoke one person at a time. | It also keeps facts and timelines, imports your ChatGPT and Claude history, keeps a version history for every page, and redacts API keys, tokens, private keys and database passwords before anything is stored. ## See it work <div align="center"> <img src="docs/assets/demo.svg" alt="An agent asks what was decided about an approach; Memrain returns cited passages from the owner's notes, and the agent answers from them." width="720"> </div> ## How it works <picture> <source media="(prefers-color-scheme: dark)" srcset="docs/assets/how-it-works-dark.svg"> <img src="docs/assets/how-it-works-light.svg" alt="Five steps: your notes, code and chats; Memrain reads them; stored in your Postgres database; search by meaning and by words; your AI agent answers with sources." width="100%"> </picture> Memrain does the remembering, not the talking (the opt-in `think` tool is the one exception). It returns ranked passages with their sources, and your agent writes the answer from them. A search costs one embedding call and no chat-model call by default. The full pipeline is in [docs/HOW-IT-WORKS.md](./docs/HOW-IT-WORKS.md). ## Get started **Self-host (free software; you pay your AWS bill).** You need an AWS account with Bedrock access, Terraform, the AWS CLI, an S3 bucket for Terraform state, and a domain on Cloudflare (or on Route53, with the Caddy option). Fork the repo to your GitHub account, then: ```bash git clone https://github.com/<your-github-username>/memrain.git && cd memrain make init # answers a few questions and writes your config make plan && make apply ``` The full walkthrough, from the tunnel token to your first search, is in [docs/QUICKSTART.md](./docs/QUICKSTART.md). **Try it on your laptop.** No servers, one embedded database. You still need AWS credentials for the embeddings. See [Try it locally](./docs/QUICKSTART.md#try-it-locally). **Hosted.** A hosted version, so you do not have to run AWS yourself, is planned. It is not available yet. ## Connect your agent Every client uses the same `https://<your-host>/mcp` address. | Client | Guide | |---|---| | Claude Code | [docs/clients/CLAUDE_CODE.md](./docs/clients/CLAUDE_CODE.md) | | Codex CLI | [docs/clients/CODEX.md](./docs/clients/CODEX.md) | | claude.ai (Pro, Max) | [docs/clients/CLAUDE_AI.md](./docs/clients/CLAUDE_AI.md) | | Claude Team, Enterprise | [docs/clients/CLAUDE_TEAM.md](./docs/clients/CLAUDE_TEAM.md) | | ChatGPT | [docs/clients/CHATGPT.md](./docs/clients/CHATGPT.md) | ## Security and privacy - **Your data stays in your AWS account.** Notes, index and database live there. Only what an agent retrieves goes to that agent's model; with the default Cloudflare Tunnel, that traffic passes through Cloudflare. No telemetry. - **Secrets are stripped on the way in.** Pasted AWS keys, API tokens, private keys and database passwords are redacted before they are stored or embedded. - **Every person has their own key.** Personal tokens and OAuth sign-in, scoped per person, each with an optional daily cap. Details in [docs/TEAM-SETUP.md](./docs/TEAM-SETUP.md). To report a vulnerability, see [SECURITY.md](./SECURITY.md). ## When Memrain is not the right fit - You do not want to run anything in AWS. A hosted version is planned; it is not available yet. - You want a chat app. Memrain is the memory behind your agent, not a chat window. ## Documentation | Doc | What is in it | |---|---| | [docs/QUICKSTART.md](./docs/QUICKSTART.md) | Install, connect, credentials, day-to-day operation | | [docs/HOW-IT-WORKS.md](./docs/HOW-IT-WORKS.md) | How search works, what it costs | | [docs/TEAM-SETUP.md](./docs/TEAM-SETUP.md) | One connector for a team, a private space per person | | [docs/DEPLOYMENT.md](./docs/DEPLOYMENT.md) | Every install and update step in detail | | [docs/CONFIGURATION.md](./docs/CONFIGURATION.md) | Every setting, quality tiers, budgets | | [ARCHITECTURE.md](./ARCHITECTURE.md) | What runs where, and the security model | | [UPGRADING.md](./UPGRADING.md) · [CHANGELOG.md](./CHANGELOG.md) | Upgrades and release history | ## Contributing Issues and pull requests are welcome. Please open an issue before anything that adds infrastructure or changes how Memrain is deployed. The local checks and test commands are in [CONTRIBUTING.md](./CONTRIBUTING.md). ## License [MIT](./LICENSE)