Ficus Tau: Workspace for Teams of AI Agents

What changed

The Ficus Tau project introduces a dedicated workspace for the coordination and management of teams of AI agents. This platform is designed to allow a group of specialized agents to work collaboratively towards a common goal, with a manager agent overseeing their activities, tracking progress, and facilitating human intervention for decisions and reviews. Users can monitor ongoing conversations, steer active agents, or allow the team to continue autonomously.

Ficus Tau supports flexible deployment options, including a self-hosted version that can be run on a user's local machine or server, and an optional Ficus Cloud service for managed hosting. The system enables squads of agents to operate across different repositories, tools, and services, accessible via a web application or a command-line interface (CLI).

Installation for local execution on macOS or Linux can be done via a curl command to download and run a setup script, which requires curl and git, and potentially unzip. The setup process installs dependencies, prompts for agent execution location, and starts the Ficus instance. Post-installation, users complete an initial sign-in, connect an AI model provider in the settings, and then create a squad with a defined goal.

Alternative installation methods include building from source by cloning the repository, running bun install, and then bun run setup. A CLI-only installation is also available for interacting with a remotely running Ficus instance. The Ficus operator skill can be installed for AI coding agents, enabling them to manage Ficus directly.

Key functionalities of Ficus Tau include:

• Delegating Work: Squads can organize tasks into work streams with defined dependencies and handoffs, supported by schedules for recurring tasks.
• Human Oversight: An Action Center, live conversations, and notifications allow users to review results, answer questions, and steer active work.
• Model and Environment Choice: Users can connect various model providers and choose execution environments for agents, ranging from local machines to Docker, VMs, or Kubernetes, offering control over isolation and infrastructure.
• Shared Knowledge: Agents retain file-based memory with keyword and vector search capabilities, preserving context across conversations.
• Tool Integration: Ficus Tau integrates with tools like GitHub, Linear, chat channels, webhooks, and app previews. The CLI and REST API support scripted workflows.
• Collaboration: Multi-user permissions manage access, and the AMTP protocol facilitates communication between agents across different Ficus instances and compatible nodes.

The project is developed using Bun, TypeScript, Hono, and React. The architecture separates the API and worker processes, with agents utilizing the same CLI and APIs available to human users. Development guides, architecture overviews, sandbox runtime details, and CLI references are available.

Ficus Tau is licensed under AGPL-3.0-only. Contributors retain copyright, with an option for Intentional Design LLC to offer contributions under additional licenses, including commercial ones, via a Contributor License Agreement (CLA).

Why it matters for builders

Ficus Tau provides a structured environment for developers to build and deploy multi-agent systems. It abstracts away much of the complexity involved in coordinating agent communication, task management, and state persistence. The platform's emphasis on modularity and extensibility allows builders to integrate custom agents, tools, and execution environments, fostering innovation in AI-driven workflows.

Practical impact

AI builders can leverage Ficus Tau to experiment with complex agent-based applications, such as automated code generation pipelines, sophisticated data analysis workflows, or customer support systems managed by specialized AI agents. The self-hosted nature allows for sensitive data processing and custom infrastructure integration. Developers can start by exploring the quick-start guide for local setup or the CLI installation for remote instance interaction. Integrating the Ficus operator skill into AI coding agents offers a direct path to automating Ficus management tasks.

Caveats and source limits

The provided source is a GitHub repository description and README. Specific details regarding performance benchmarks, pricing for Ficus Cloud, or a definitive release date beyond "fresh release" are not available. The excerpt mentions "4 AI signals, 6 developer signals" which are internal metrics and not independently verifiable. The license is AGPL-3.0-only, which may have implications for commercial use and distribution.