Your AI coding tool forgets everything between sessions, and you pay full price for the routine work. aimee is a local server that fixes both. It holds memory and a map of your code across tools, routes cheap work to cheap models, and enforces guardrails the model cannot write past.
Point any tool at it. Your context follows you between them.
Two services:
- aimee-server assists one human. Sessions, tools, credentials, delegates, workflows.
- aimee-kb serves a corpus, team, or company. Durable knowledge, code indexes, retrieval, curation.
The aimee CLI is a thin client for Linux, macOS, and Windows. Core services are C, the workflow
control plane and browser are Go, and nothing phones home.
- Memory that survives the session. The curator extracts facts, joins evidence, catches contradictions, and lets stale detail decay. See Knowledge.
- Your code as a graph. Symbols, callers, imports, and cross-repo dependencies feed search and blast-radius checks. See Code intelligence.
- Delegates that cut the bill. Send review, diagnosis, and routine implementation to the cheapest model that fits the role. See Delegates.
- Guardrails before execution. Secret paths, unsafe writes, worktree escapes, and untrusted MCP packages are checked first, and delegate sandboxes run with no network and no credentials. See Security.
- One bus, one audit trail. Every governed action, memory write, guardrail decision and vault read crosses one sequenced tap into a WORM ledger. See Event bus.
- Any provider. OpenAI, Anthropic, Gemini, Mistral, Bedrock, and local OpenAI-compatible servers pass through one internal request format. Switch models without switching tools.
git clone https://github.com/RakuenSoftware/aimee.git
cd aimee
docker compose -f compose.server-managed.yaml up -d
docker compose -f compose.server-managed.yaml logs aimee-serverThe log prints a generated, one-time dashboard login. Open https://localhost:8443, sign in, and
the setup wizard covers the account, provider, knowledge base, deployment, and workspaces. After the
numbered steps, the summary can start aimee-kb with PostgreSQL 18, pgvector, and pgvectorscale
inside the container.
The managed compose file mounts the Docker socket, which gives aimee-server control of the host Docker daemon. Use the split stack if you do not want that.
The Quickstart has the rest: choosing your own login, the split stack, thin client enrollment, and what to check when it does not work.
Start at the documentation index.
| Document | Use it for |
|---|---|
| Quickstart | Install, enroll, verify. |
| What's new | Everything 0.3.0 changed, and what it removed. |
| Upgrading | Move from v0.2.192. One-way, so read it first. |
| Manual | Day-to-day use and operations. |
| Architecture | Processes, storage, trust, request flow. |
| Deployment | Managed, split, external DB2, backup, hardening. |
| Command reference | Every CLI command. Generated from source. |
| Configuration reference | Every config key and variable. Generated from source. |
| Server API | /v1 transport, auth, compatibility. |
| Feature status | What works, what is gated, what was removed. |
| Troubleshooting | Diagnose the first broken boundary. |
Questions and discussion: https://discord.gg/FjGjvcgAqz.
Copyright (C) 2026 The aimee authors. Licensed under the GNU AGPL v3.0. See LICENSE and NOTICE.
If the AGPL does not suit you, other terms can be discussed. Contact jbailes@gmail.com. Bundled components and generated SDKs may use different licenses; NOTICE lists them.