Agents That Remember
Who You Are
Most agent frameworks treat every conversation as day one. NarraNexus gives agents persistent memory, social identity, and evolving expertise — across sessions, topics, and teams.
The problems we solve
“Agents forget everything between sessions”
Narrative Memory
Conversations are organized into topic-based storylines that persist for weeks. Your agent picks up right where you left off — even if the last message was 3 weeks ago on a different topic.
“Agents don’t know who they’re talking to”
Social Intelligence
Agents build a graph of everyone they interact with — names, roles, expertise, communication style. They adapt how they work based on who’s asking.
“Agent capabilities are monolithic and rigid”
Modular Architecture
Hot-swap capability modules without touching core logic. Add memory, jobs, skills, or agent-to-agent communication as independent plugins.
What makes NarraNexus different
Narrative Memory
Every topic becomes a persistent Narrative with its own history, active modules, and summary. The agent routes new messages to the right storyline automatically — or creates a new one when the topic is genuinely new.
Learn more →Social Graph
Identity, expertise, contact details, and communication personas — all learned from conversations and updated continuously. The agent remembers who prefers bullet points, who needs formal reports, and who hasn’t been contacted in a while.
Learn more →Autonomous Work
Schedule one-off reminders, recurring checks, or ongoing monitoring tasks. Jobs run in the background with full context and report results through chat when they complete.
Learn more →Extensible by Design
Modules are independent and hot-swappable. Skills install from a marketplace. LLM providers swap without code changes. Build with Claude, OpenAI, or Gemini — or switch between them.
Learn more →System Overview
A layered pipeline from user interface to persistent storage.
See it in action
Topic-Aware Conversations
- 1Messages automatically routed to the right topic storyline
- 23-stage selection: continuity detection → semantic search → agent decision
- 38 default Narratives per user — from greetings to deep technical discussions
- 4Cross-topic awareness: the agent knows what you discussed elsewhere
Who is it for
Personal Assistant
An agent that remembers your preferences, manages reminders, and adapts to how you like to communicate — across days, weeks, and topics.
Team Coordinator
Multi-agent systems where each agent has a specialty — research, scheduling, code review — and they coordinate through the MessageBus.
Developers
Build agents with persistent memory and social awareness in hours, not weeks. Modular architecture means you extend, not rewrite.
Researchers
Experiment with narrative-driven agent architectures. Study how topic-based memory and social graphs affect agent behavior over time.
Try it in 60 seconds
Three ways to get started, from zero-install to full local setup.
No installation. Open in your browser, create an account, and start chatting with your agent immediately.
Open NarraNexus WebDownload the macOS app. Bundled Python, auto-starts all services. No terminal required.
Download (macOS) →Full local setup with tmux, 7 services, and hot-reload for development.
git clone ...NarraNexus && bash run.shWhere we are going
- Core agent runtime
- Matrix protocol integration
- Basic memory system
- Docker deployment
- Narrative memory engine
- Multi-LLM support
- Skill marketplace
- Cost tracking dashboard
- MCP server integration
- Agent-to-agent workflows
- Advanced RAG pipeline
- Community plugin system