What it does

NarraNexus is a platform for keeping a standing team of AI agents rather than disposable assistants. Each agent holds a lasting identity, a memory that develops over time, a set of tools of its own and stable relationships with people and other agents. Agents can @mention each other over a protocol called MessageBus, open rooms, talk in groups, divide up work and pass tasks along. Memory is organized into storylines that an agent can continue across sessions. When a conclusion is overturned it is marked invalid rather than deleted, so the history is kept. Memories are scoped to a single agent, a user, a storyline, a team or everyone, and sharing between agents goes through governed channels. Each agent can be given its own tools, Skills and MCP services, added without changing global code. You can start in the cloud with a free trial quota, install a macOS desktop app that bundles its runtime, or run from source with a shell script. Running locally or with the desktop app requires you to supply an LLM API key. The project is Apache-2.0 licensed and ships templates such as a financial morning briefing, a KOL assistant and a PM bridge bot. The page is open about limits: agents need corrections and feedback and improve with use.

Capabilities

Agents with persistent identity and long-term narrative memory
Agent-to-agent messaging, rooms and task hand-offs through MessageBus
Memory scoped per agent, user, storyline, team or globally
Per-agent Skills and MCP tool sets
Ready-made templates such as a financial morning briefing
Cloud sign-up, a macOS desktop app, or run from source

When to use it

Running a standing team of agents with distinct roles such as research, product and content
Long projects where agents must keep context across days
Continuously tracking markets or competitors with a research agent
Client communication that needs separate internal and client-shared knowledge