Most agent frameworks focus on making agents smarter. NarraNexus focuses on making agents connected.
An agent in isolation is a tool. An agent with memory, identity, relationships, and goals becomes a participant in a nexus — a network where intelligence is a collective property, not just a model property.
NarraNexus provides the infrastructure for this: persistent memory, relationship-aware context, task scheduling, modular capabilities, and agent-to-agent communication.
Agents that remember — across sessions, conversations, and relationships.
NarraNexus agents carry context across conversations through long-term memory, event memory, and relationship-aware retrieval. They continue from past interactions instead of starting over every time.
Every capability is a hot-swappable module.
Core capabilities such as Memory, Awareness, Chat, RAG, Jobs, Skills, Social Network, and Matrix run as independent modules. Each module manages its own tools, data, and lifecycle, making the system easy to extend or customize.
Built for collaboration, not just conversation.
Agents can communicate through Matrix-based messaging and use MCP tools to coordinate with other agents, external tools, and background workflows.
Try NarraNexus instantly in the browser — no install needed.
Native desktop app with auto-updater.
Download Latest Release → — choose the file ending with
.dmg.
| Dependency | Install |
|---|---|
| Node.js (v20+) | Install via nvm (recommended): curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.1/install.sh | bash && nvm install 20 |
| uv | curl -LsSf https://astral.sh/uv/install.sh | sh |
git clone https://github.com/NetMindAI-Open/NarraNexus.git
cd NarraNexus
bash run.shTip
The script auto-detects your OS (Linux / macOS / Windows WSL2) and handles the rest of the dependencies. If either dependency is missing, run.sh will print the install command and exit. Install it, then re-run.
Once setup completes:
- Open
http://localhost:5173in your browser- Choose LOCAL or CLOUD (Coming soon) mode to create an account and log in
- Click SETTING on the left panel to set up the API key — see LLM Provider Configuration
- Start chatting!
- Open
http://localhost:8000/docsfor API Docs
Setup complete — ready to open the interface
Note
For more details, see the installation instructions in the docs.
The agent uses three functional LLM slots:
| Slot | Protocol | Purpose |
|---|---|---|
| Agent | Anthropic | Core reasoning — powers thinking, tool use, and multi-turn conversations |
| Embedding | OpenAI | Converts text to vectors for narrative matching and semantic search |
| Helper LLM | OpenAI | Lightweight tasks — entity extraction, summarization, module decisions |
Configuration is done in two steps:
- Add a provider
- Assign a model to each slot
Use Quick Add — Preset Provider to select a provider and paste your API key. Preset providers such as NetMind.AI Power can automatically create both Anthropic-compatible and OpenAI-compatible endpoints from one API key.
You can also configure:
| Option | What you need | Result |
|---|---|---|
| NetMind.AI Power | One API key | Creates both Anthropic and OpenAI endpoints automatically |
| OpenRouter / Yunwu | One API key | Adds supported endpoints and available models |
| Claude Code Login | Claude Code CLI login | Enables Claude models for the Agent slot through OAuth |
| Custom Anthropic | Compatible URL and API key | Adds a custom Anthropic endpoint |
| Custom OpenAI | Compatible URL and API key | Adds a custom OpenAI endpoint |
Use Update Available Models to refresh the default model list for preset providers. Existing model entries are kept, and only missing models are added.
After adding providers, go to Model Assignment and select a provider and model for each slot:
| Slot | Example |
|---|---|
| Agent | NetMind Anthropic + DeepSeek V4 Pro(more available), or Claude Code + Claude model |
| Embedding | NetMind OpenAI + embedding model |
| Helper LLM | NetMind OpenAI + DeepSeek V4 Pro(more available) |
All three slots must be configured before the agent can work.
Note
For updating LLM configuration, click Setting see the installation instructions in the docs.
| Feature | Description |
|---|---|
| Narrative Memory | Conversations routed into semantic storylines, retrieved by topic similarity across sessions |
| Hot-Swappable Modules | Standalone capabilities (chat, social graph, RAG, jobs, skills) with their own DB, tools, and hooks |
| Inter-Agent Communication | Agents coordinate via Matrix protocol — rooms, messages, @mentions, group chats |
| Skill Marketplace | Browse and install skills from ClawHub via natural language |
| Social Network | Entity graph tracking people, relationships, expertise, and interaction history |
| Job Scheduling | One-shot, cron, periodic, and continuous tasks with dependency DAGs |
| RAG Knowledge Base | Document indexing and semantic retrieval via Gemini File Search |
| Long-term Memory | Episodic memory powered by EverMemOS (MongoDB + Elasticsearch + Milvus) |
| Cost Tracking | Real-time metering of every LLM call with per-model cost breakdowns |
| Execution Transparency | Every pipeline step visible in real time — what the agent decided, why, and what changed |
| Multi-LLM Support | Claude, OpenAI, and Gemini via unified adapter layer |
| Desktop App | Native desktop application with auto-updater and one-click service orchestration |
NarraNexus in action
NarraNexus's long-term memory system is built on EverMemOS, a self-organizing memory operating system for structured long-horizon reasoning. We thank the EverMemOS team for their foundational work.
Chuanrui Hu, Xingze Gao, Zuyi Zhou, Dannong Xu, Yi Bai, Xintong Li, Hui Zhang, Tong Li, Chong Zhang, Lidong Bing, Yafeng Deng. EverMemOS: A Self-Organizing Memory Operating System for Structured Long-Horizon Reasoning. arXiv:2601.02163, 2026. [Paper]
If you find NarraNexus useful, please cite it as:
@software{narranexus2026,
title = {NarraNexus: A Framework for Building Nexuses of Agents},
author = {NetMind.AI},
year = {2026},
url = {https://github.com/NetMindAI-Open/NarraNexus},
license = {CC-BY-NC-4.0}
}