Show HN: Manifold is a platform for workflow automation using AI assistants

Manifold is a powerful platform designed for workflow automation using AI models. It supports text generation, image generation, and retrieval-augmented generation, integrating seamlessly with popular AI endpoints including OpenAI, llama.cpp, Apple's MLX LM, Google Gemini, Anthropic Claude, ComfyUI, and MFlux. Additionally, Manifold provides robust semantic search capabilities using PGVector combined with the SEFII (Semantic Embedding Forest with Inverted Index) engine.

Note: Manifold is under active development. The platform is not intended for production use until otherwise noted.

Run parallel inference across multiple providers.

Manifold implements an MCP client and a ReAct Agent for full "autopilot" task execution. MCP servers can be configured as Docker containers that mount the same local:container paths for secure tool execution.

Or use the MCP node to list and execute tools as part of your workflows.

Manifold supports image generation via ComfyUI in Windows, Linux and MacOS. Manifold also supports MLXFlux image generation on MacOS only, no external service required.

For now, Manifold supports OpenAI and supported local vision models via llama.cpp such as Gemma-3.

Manifold features a powerful semantic and keyword document ingestion and retrieval engine. Ingest files, entire folders, and even LLM output. Retrieve relevant information for your AI assistant to use as reference.

Manifold provides nodes to control the execution order of your workflows.

Run concurrent agents. Jump to nodes. Execute for each loops on assistant outputs. Combine context. Publish and subscribe to topics to teleport context around your workflows. Be creative!

Ensure the following software is installed before proceeding:

• Chrome Browser: Required for web tools. Used as a headless browser and managed by Manifold. WebGPU support should be enabled for text-to-speech to work.
• Python: Version 3.10 or newer (Download).
• Docker: Recommended for easy setup of PGVector (Download).

For development, you'll also need:

• Go: Version 1.21 or newer (Download).
• Node.js: Version 20 managed via nvm (Installation Guide).

The easiest way to get started with Manifold is to download a pre-built binary from the releases page.

1. Download the appropriate binary for your platform:

   • macOS: manifold-darwin-universal.zip (Universal binary for both Intel and Apple Silicon)
   • Linux: manifold-linux-amd64.zip or manifold-linux-arm64.zip

2. Extract the zip file and navigate to the extracted directory.

3. Create a config.yaml file in the same directory as the binary (a template config.yaml.example is included).

4. Run the binary:

   # On macOS/Linux $ chmod +x manifold-* $ ./manifold-*

After cloning the repository, initialize and update the git submodules:

This will fetch the required dependencies:

• llama.cpp for local model inference
• pgvector for vector similarity search in PostgreSQL

Manifold will automatically manage the lifecycle of the PGVector container using Docker. Ensure Docker is installed and running on your system.

• Follow the official ComfyUI installation guide.
• No extra configuration needed; Manifold connects via proxy.

• Follow the MFlux installation guide.

Use the provided config.yaml.example template to create a new config.yaml file. This file must be placed in the same path as the main.go file if running in development mode, or in the same path as the manifold binary if you build the project.

Ensure to update the values to match your environment.

For development it is not necessary to build the application. See development notes at the bottom of this guide.

Execute the following commands:

This sequence will:

• Switch Node.js to version 20.
• Build frontend assets.
• Compile the Go backend, generating the executable.
• Launch Manifold from the dist directory.

Upon first execution, Manifold creates necessary directories and files (e.g., data).

Note that Manifold builds the frontend and embeds it in its binary. When building the application, the frontend is not a separate web server.

• On first boot, the application will take longer as it downloads the required models for completions, embeddings, and reranker services.
• The application defaults to a single node instance configuration, managing the lifecycle of services using the llama-server backend and bootstrapping PGVector.
• Services can be configured to run on remote hosts to alleviate load on a single host, but users must manage the lifecycle of remote services manually.

Create or update your configuration based on the provided config.yaml.example in the repository. Manifold uses a flexible configuration system that supports both YAML files and environment variables.

Ensure you rename the file or create a new config.yaml with your unique configuration, and the file is present in the same path as the Manifold binary.

Crucial Points:

• Update database credentials (myuser, changeme) according to your PGVector setup.
• When single_node_instance is enabled, Manifold auto-manages the lifecycle of llama-server instances.
• When using external API services (OpenAI, Claude, etc.), provide the corresponding API keys.

Launch your browser and navigate to:

Replace the host configuration if you customized it in config.yaml.

When you first access Manifold, use these default credentials to log in:

⚠️ IMPORTANT SECURITY WARNING: These are publicly known default credentials. Immediately after logging in, change your password by clicking on your account name in the top right corner and selecting "Change Password".

Manifold is compatible with OpenAI-compatible endpoints:

• llama.cpp Server
• Apple MLX LM Server

Manifold provides a Docker image for the Model Context Protocol (MCP) server component, allowing you to run the MCP server in containerized environments.

To build the Docker image:

Once built, you can configure the MCP server in your config.yaml:

• Port Conflict: If port 8080 is occupied, either terminate conflicting processes or choose a new port in config.yaml.
• PGVector Connectivity: Confirm your database.connection_string matches PGVector container credentials.
• Missing Config File: Ensure config.yaml exists in the correct directory. Manifold will not launch without it.

Manifold welcomes contributions! Check the open issues for tasks and feel free to submit pull requests.