Brand-new release of memory system api to power various agent applications

🚀 MIRIX 0.1.6 — Pure Memory System API for Any Agent

This release marks a major evolution of MIRIX.

Starting from v0.1.6, MIRIX becomes a dedicated, plug-and-play advanced memory system that any agent can integrate with — rather than a bundled desktop personal assistant.

With this release, MIRIX now provides:

• ⚙️ Unified Memory API
• 🧠 Meta-agent & multi-memory architecture
• ☁️ Hosted Cloud Memory Service
• 📊 Web Dashboard for visualization, debugging & control
• ⚡ Ultra-fast retrieval with hybrid search
• 🔍 Optional embedding search for deeper reasoning
• 🔐 Secure API key–based access

---

✨ What’s New

✅ MIRIX as a Standalone Memory Platform

MIRIX is now designed to act as the long-term memory layer for:

• LLM agents
• desktop assistants
• chatbots
• automation systems
• multi-agent systems

Just install and start using:

pip install mirix==0.1.6

---

☁️ MIRIX Cloud + Dashboard

We now provide:

• Hosted Cloud Memory Service (no infrastructure needed)
• Memory Dashboard
  • inspect stored memories
  • search & debug
  • monitor usage
  • manage API keys

Sign up at:
👉 https://app.mirix.io

---

🔄 Branches & Migration

• main (≥ v0.1.6) → Pure Memory System (API + Cloud)
• desktop-agent branch → Legacy Desktop Personal Assistant
  (deprecated, but still available)

The previous desktop personal assistant (frontend + backend) is no longer shipped on main.

If you still need the old integrated desktop app:
👉 switch to desktop-agent branch

---

🧨 Breaking Changes

• Desktop personal assistant removed from main
• API-first architecture introduced
• Cloud usage requires API key

---

🧪 Example Usage

Add Memory

client.add(
    messages=[
        {"role": "user", "content": [{"type": "text", "text": "We discussed MirixDB indexing performance"}]},
        {"role": "assistant", "content": [{"type": "text", "text": "Noted, I will remember that"}]},
    ],
)

Retrieve Conversational Memory

memories = client.retrieve_with_conversation(
    messages=[
        {"role": "user", "content":[{"type":"text","text":"What did we discuss recently?"}]},
    ],
)
print(memories)

Deep Embedding Search

client.search(
    memory_type="episodic",
    search_field="summary",
    search_method="embedding"
)

---

❤️ Thank You

Thanks to everyone supporting MIRIX — this upgrade enables MIRIX to power agents of all kinds while scaling to real-world workloads. Much more is coming 😊

If you’re building with MIRIX, we’d love to hear what you create!

Add Multi-User Support, Gmail MCP, Web Search, Better Memorization Triggering

🚀 Update Log

New Features

• Multi-User Support : Added support for multiple users in both the SDK and Desktop App. 📖 See documentation
• Gmail MCP Tool: Integrated Gmail as an MCP tool for seamless email connectivity.
• Web Search: Introduced built-in Web Search for more dynamic and up-to-date responses.
• SDK: New PYPI package at https://pypi.org/project/mirix/

Improvements

• Enhanced Memorizing Mechanism: Previous complaints about the chat agent not saving information into memory have been fully resolved. Memory retention now works reliably and consistently.

Fixes

• Summarization issues with Azure
• Customized model configuration problems
• Other small stability improvements

Application-Specific Monitoring & Python SDK

Release Notes

1. Application-Specific Monitoring: Implemented selective application monitoring functionality, allowing users to restrict agent visibility to specific applications rather than the entire screen.

2. Python SDK Released: Introduced official Python SDK for enhanced integration capabilities. Available at https://pypi.org/project/mirix/

3. System Improvements:

   • Modified backend port to 47283 to enhance system stability and prevent conflicts with other applications
   • Added support for OpenAI models as memory managers (note: may increase latency)
   • Implemented custom API key configuration for third-party models including Qwen3-32B
   • Enhanced UI functionality and resolved various minor issues in the frontend

support vllm served models

1. vLLM Integration for Model Serving: MIRIX now supports serving models using vLLM. Developers can register their models through the UI by following the configuration template provided in configs/mirix_custom_model.yaml. For detailed instructions, refer to our documentation at https://docs.mirix.io/user-guide/custom-models/.

2. Enhanced Chat Agent Behavior: The chat agent has been upgraded to allow direct updates to the memory system, even when screen monitoring is disabled.

The initial version

MIRIX is the initial release of the personal assistant with our most advanced memory system that can see your screen, record your activities and understand you.

Features

• Separated backend and frontend architecture for developers.
• Easy installation via .dmg file.
• Allow multi-modal conversation. User can drag the images into the chatting window.
• Can monitor the screen via screenshots and build your personal memory.

Installation

1. Download the .dmg file below.

2. Open it and drag MIRIX into your Applications folder.

3. Launch the app.

4. Because this app is not notarized by Apple, you may see a security warning when launching it. To allow the app to run, you can either:

   • Right-click (Control-click) the app in Applications and choose "Open," then confirm in the dialog.

   or

   • Run this command in Terminal to remove the quarantine attribute:

     xattr -d com.apple.quarantine /Applications/MIRIX.app
     

   This is expected behavior for open-source software distributed outside the App Store. The app is built directly from the code in this repository.

Known Issues

• The app may not support dark mode yet.
• There might be unknown bugs so we greatly appreciate it if you can report the bugs to us so we can improve!

Feedback

If you encounter problems, please open an issue on this repository.