Repolode
Mirix-AI

Mirix-AI/MIRIX

Multi-agent personal assistant that watches on-screen activity and builds structured, locally stored memories to answer questions about your digital work.

Stars
3,456
Forks
273
Open issues
46
Last commit
14 days ago

What is it?

Mirix is a personal AI assistant that captures screen activity and conversations, then consolidates them into structured long-term memory. Six memory components (core, episodic, semantic, procedural, resource, and knowledge vault) are each managed by a dedicated agent, so information is organized by type rather than stored as one flat history.

The project runs a backend and dashboard via Docker, exposes a REST API on port 8531, and provides a Python client (mirix-client) for adding messages and retrieving memories with conversation context. It supports text, images, voice, and screen captures as input. Search is built on PostgreSQL with BM25 full-text search and vector similarity.

Mirix includes an auto-dream endpoint that reviews existing memories, merges duplicates, and resolves conflicting entries, with a dry-run option to preview changes. Procedural memory can distill skills from tool calls and errors across recent sessions. Existing databases need manual migration scripts when upgrading.

Key features

  • 01Six specialized memory types, each managed by a dedicated agent
  • 02Continuous screen capture consolidated into structured memories
  • 03Local storage of long-term data with user-controlled privacy settings
  • 04PostgreSQL-based BM25 full-text search with vector similarity
  • 05Accepts text, image, voice, and screen capture input
  • 06Auto-dream endpoint for merging and cleaning up memories

Use cases

  • Recalling what you worked on or discussed across recent days
  • Building a personal knowledge base from screen activity and chats
  • Distilling reusable procedures from tool errors and fixes in agent workflows
  • Experimenting with multi-agent memory architectures for LLM assistants

Quick start

  1. 1

    Start the backend and dashboard with Docker

    docker compose up -d --pull always
  2. 2

    Create an API key in the dashboard and set it as the MIRIX_API_KEY environment variable

  3. 3

    Install the Python client

    pip install mirix-client

Things to know

  • Requires a configured LLM and embedding provider, such as a Google AI API key, to run.
  • Stores screen-capture-derived memories, which raises privacy considerations for shared machines.
  • Self-hosting needs Docker and a PostgreSQL-backed setup, and existing databases need manual migrations on upgrade.

Topics

FAQ

What is MIRIX?
Multi-agent personal assistant that watches on-screen activity and builds structured, locally stored memories to answer questions about your digital work. Mirix is a personal AI assistant that captures screen activity and conversations, then consolidates them into structured long-term memory. Six memory components (core, episodic, semantic, procedural, resource, and knowledge vault) are each managed by a dedicated agent, so information is organized by type rather than stored as one flat history.
Is MIRIX free to use? What license does it have?
MIRIX is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is MIRIX still maintained?
Actively maintained. The last commit to MIRIX was on Oct 8, 2026.
What can I use MIRIX for?
Recalling what you worked on or discussed across recent days Building a personal knowledge base from screen activity and chats Distilling reusable procedures from tool errors and fixes in agent workflows Experimenting with multi-agent memory architectures for LLM assistants
How do I install MIRIX?
Start the backend and dashboard with Docker: docker compose up -d --pull always Create an API key in the dashboard and set it as the MIRIX_API_KEY environment variable Install the Python client: pip install mirix-client
Where can I get MIRIX?
The source code and installation instructions are on GitHub: https://github.com/Mirix-AI/MIRIX

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects