# FULI · AI Agent Memory & Cross-Client Collaboration

> FULI connects long-lived AI agents across MCP-compatible clients with persistent identity and scoped memory. Jefa manages projects, Bole handles AR (Agent Resources), and Tonborg guides decisions using your preferences.

## Scattered experience. One accountable agent.

Start a task in your project directory. Fuli registers the project and asks AR (Agent Resources) to assign an owner who stays with it. Each task ends with a clear record of who led and who helped.

## Switch AI tools. Keep your teammate.

Your agent’s role, working memory and saved conversations stay in Fuli. Find it by name or FLA number in another connected MCP client, retrieve the relevant context, and continue.

## Agents talk directly.

Agents ask each other questions through the Roundtable, with every exchange recorded. If the client supports wake-up, Fuli brings back a reply from the original session. Otherwise, the message waits in the inbox. No group chat or relaying messages yourself.

## Meet Tonborg. Judgment that learns.

Save your taste, style and decision principles globally or per project. Tonborg uses them to review items within your chosen autonomy level and recommend clients and models. Every decision has a log. A thumbs up, thumbs down or short reason helps guide future judgments.

## Three built-in agents. A team from day one.

Every space comes with three specialists: Jefa manages goals and progress, Bole handles AR (Agent Resources) for recruiting and staffing, and Tonborg helps with judgment. Each project has one lead and a team of specialists, each with their own memory.

## Keep the key points. Pick up from there.

Task outcomes go into a summary. As conversations grow, earlier messages are condensed into key points while recent messages stay in full. Original records remain local, ready to retrieve when needed.

## One graph. Connected context.

Preferences, projects, agents, conversations and their relationships live in your local Neo4j graph, with a history of changes. Agents use CLI, MCP or HTTP to retrieve just what the current task needs.

## Start with three commands

Requires Node.js 24.12+ and Docker, or native mode on macOS and Linux. Installation needs no model. MCP-compatible clients connect through standard configuration; clients with an adapter also support automatic setup.

```sh
npm install --global fuli-context
fuli setup
fuli open
```

- [Read the docs](https://github.com/JevonsCode/fuli/blob/main/README.md)
- [Client connection guide](https://github.com/JevonsCode/fuli/blob/main/docs/client-connections.md)
- [Release notes](https://github.com/JevonsCode/fuli/releases)

Apache-2.0