Mimwell is a mind palace you run yourself: an open-source knowledge base and context store. It compiles your documents, decisions, plans and records into one graph and search index. A portal serves that graph to people, and a retrieval agent answers questions from it.
The name comes from Mímir's well, the spring under the world tree whose water held wisdom and memory. Mimwell does the same job for a team: it keeps what the team learned somewhere anyone can draw from.
What is a mind palace?¶
It is the oldest memory trick there is. Greek and Roman orators placed each thing they wanted to remember in a room of an imagined building, then walked through it to recall. Mimwell builds that palace for a team and its AI agents: every document, decision and plan gets a place, and anyone can walk in and find it.
Build one for¶
- Companies and organizations: strategy, decisions and how things work.
- Departments and teams: plans, owners, open questions and handoffs.
- Projects: scope, commitments, risks and what changed.
- Courses and topics: a curated body of knowledge people can explore.
- AI development: grounded context for coding agents and AI apps.
Why you want one¶
- Nothing important gets lost. Decisions keep their reasons and their sources, so the next person finds the why, not only the what.
- One place to ask. Search and chat answer from your own material, with citations back to the source.
- Faster onboarding. New people read the palace instead of interviewing everyone.
Built for AI development¶
- A context store your agents can read. The same graph and search index the portal serves are there for coding agents and AI apps, so they work from your real decisions and docs instead of guessing.
- Grounded answers. The retrieval agent answers from the palace's own content and cites it, which keeps answers tied to what you wrote.
- Scoped access. Per-path read rules apply to pages, search and agent answers alike, and fail closed.
- Reproducible context. The build is deterministic, so the same sources always give agents the same context.
- Curated, not dumped. Suggested connections land in a review queue, and nothing becomes part of the palace until someone accepts it.
What you get¶
- A compiled brain. Markdown documents and structured records build into a knowledge graph, a search pack and a manifest. The build is deterministic, so the same inputs always produce the same outputs.
- A portal. A static site served by a Cloudflare Worker, with an Easy view for readers and an Advanced view for people who want every record kind, the graph and the sources.
- A retrieval agent. Chat that answers from the brain's own content and cites its sources.
- Your brand. Name, colors, fonts and marks come from one config file.
- Access control. Optional per-path read scoping behind Cloudflare Access, enforced on every page, search result and agent answer. It fails closed.
- Brain apps. Small, sandboxed apps that run beside the portal.
- Sensemaking. An opt-in workbench for concepts, evidence, assessments and strategic maps.
- Many sources. Documents, wiki pages, structured registers, issue and activity snapshots, and database schemas. See Data sources.
- Your choice of storage. JSON out of the box, an optional local semantic index, and blob stores on disk, in a database, on S3-compatible storage or Azure. See Storage.
- A context API for agents. Authenticated, policy-filtered and pinned context for coding agents and AI apps. See Context API.
- Reviewed curation. Suggestions wait for a reviewer, and decisions keep their full lineage. See Curation.
Where to go next¶
- Installation: what you need on your machine.
- Quick start: build and preview your first brain.
- Concepts: how a brain is put together.
- Configuration: make it yours.
- Deploying: put it on your own domain.
- Data sources: everything a brain can read.
- Context API: give your agents grounded context.
Mimwell is open source under the Apache License 2.0, built by CONFLICT.