CremindWhat is Cremind?
Cremind is an open-source personal AI agent you run yourself, with a unified tool registry spanning built-in tools, Agent Skills, MCP, and A2A.
Cremind is an open-source personal AI agent that runs on your own machine or server — not a hosted service you rent. You bring your own LLM key (Anthropic, OpenAI, or Groq), and Cremind gives the model a runtime, a tool plane, a web UI, and a CLI bundled into one package. It plans and reasons with a model, then gets real work done by calling tools, running scripts, and reacting to events as they happen.
Cremind is early — version 0.0.1. It is open source under the MIT license and community-driven, so expect rapid change and rough edges. The docs describe how it works today.
A personal agent you own
Most agent platforms are SaaS: your data flows through someone else's servers and your automations live behind their dashboard. Cremind inverts that. You install it, it runs locally (or on a server you control), and your conversations, keys, and indexed documents stay where you put them.
A single install can host more than one assistant at once through profiles. Each profile isolates its own:
- persona and system prompt,
- LLM provider keys,
- installed skills,
- browser profiles,
- conversation history.
So the same machine can run a work assistant, a coding assistant, and a home assistant side by side without leaking context between them.
The unified tool registry
The core idea in Cremind is that everything the agent can do is exposed through one registry. Whether a capability is built in, packaged as a skill, provided by an external server, or delegated to another agent, the planning model sees a single, consistent catalog of tools. Four sources feed that registry:
| Source | What it is |
|---|---|
| Built-in tools | First-party capabilities shipped with Cremind: a file browser, a terminal, document RAG over a watched folder, and Playwright-driven browser automation. |
| Agent Skills | Packaged units of capability — a SKILL.md plus a few scripts — that can also ship a long-running event listener. |
| MCP | Any Model Context Protocol server you point Cremind at, plugged in as tools. |
| A2A | Other agents reached over the open Agent-to-Agent protocol, so Cremind can delegate to or compose with them. |
Because all four land in the same place, you can mix them freely in a single task — read a file with a built-in tool, enrich it via an MCP server, and hand a sub-task to another agent over A2A — without the model needing to know where each capability came from.
Two open protocols
Cremind speaks MCP for pluggable tools and A2A for pluggable agents. Both are open standards, so you are not locked into a single vendor's ecosystem.
Built-in capabilities
Out of the box, the agent can browse and edit files, run terminal commands, search your documents, and drive a real browser with Playwright. Drop files into a watched folder and Cremind indexes them for retrieval, so the agent can answer questions grounded in your own material. Storage starts on SQLite and can scale to Postgres, Qdrant, or ChromaDB when you need it — chosen per service in the setup wizard.
Where to go next
Why Cremind
The advantages that set Cremind apart from other agent platforms.
Architecture overview
How the runtime, tool plane, and event system fit together.
Quickstart
Install in one command and finish setup in a few minutes.
Agent Skills
Package capabilities — and event listeners — as installable skills.