intelligencedev/manifold

What it solves

Manifold is an experimental platform designed for long-horizon workflow automation. It moves beyond simple one-shot prompts to enable teams of AI assistants (specialists) to collaborate on complex, multi-step objectives that may take hours to complete.

How it works

The platform provides a workspace where users can define "specialists" (AI agents) and organize them into projects. It integrates with various LLMs (OpenAI, Google, Anthropic, and local models via llama.cpp or vLLM) and uses the Model Context Protocol (MCP) to extend agent capabilities with external tools.

Key components include:

• Workflow Editor: A visual flow editor where MCP tools are converted into nodes, allowing users to build complex agentic sequences that can then be reused as tools themselves.
• Pulse: A scheduler for executing agent tasks at specific intervals or dates.
• Specialist Registry: A system for configuring and managing a team of expert agents.
• Projects: Isolated workspaces with their own root paths and skill sets.
• Playground: A tool for versioning prompts and running experiments to test agent behavior.

Who it’s for

It is intended for developers and AI researchers who want to build and automate complex, multi-step AI workflows using a team of specialized agents.

Highlights

• Visual Workflow Design: Create agent workflows visually and save them as tools for other agents to use.
• MCP Support: Native integration with the Model Context Protocol for easy tool expansion.
• Multi-Model Support: Compatible with frontier models from major providers and self-hosted open-weight models.
• Multimodal Capabilities: Supports image generation via OpenAI, Google, and ComfyUI.
• Scheduled Automation: Ability to schedule tasks via Pulse and send results to external services like Matrix.