A modular multi-agent memory system that augments any LLM. Specialized agents manage six memory types (Core, Episodic, Semantic, Procedural, Resource, Knowledge Vault) under a coordinator that orchestrates updates and retrieval. Ships a desktop app that builds a personal memory base from on-screen activity.
- Storage
- Typed memory stores for the six memory categories; local storage for the packaged app to keep data on-device.
- Retrieval
- A coordinator routes queries to the relevant memory-type agents and consolidates results; operates as an external system, not a model plugin.
- Self-host
- Self-host: moderate
- License
- Apache-2.0
- Pricing
- Open source (Apache-2.0), free to self-host · Free / OSS
- GitHub stars
- 3,549
- Last release
- 2025-12-25
- Last commit
- 2026-06-20
- First catalogued
- 2026-06-28
Strengths
- Multi-agent design with six distinct memory types
- Model-agnostic — augments any LLM as an external system
- Ships a working personal-assistant app with local storage
Watch out
- No tagged GitHub releases yet; versioning/stability unclear
- A hosted offering may exist (mirix.io); managed pricing not confirmed
- Self-reported LoCoMo/ScreenshotVQA numbers exist — route to harvest-benchmarks, do not treat as neutral here
Best for
- Personal assistants needing multimodal, screen-aware long-term memory
Benchmark results
| Benchmark | Value | Backbone | Trust | Source |
|---|---|---|---|---|
| locomo | 85.38 accuracy | gpt-4.1-mini | Self-reported | MIRIX (Wang & Chen) ↗ |
| longmemeval | 43.49 accuracy | GPT-4o-mini | Independent | MemOS (MemTensor) — competitor re-run ↗ |
Sources
- https://github.com/Mirix-AI/MIRIX (vendor)
- MIRIX: Multi-Agent Memory System for LLM-Based Agents (paper)
Last verified 2026-06-28 · updated by manual-refresh