ix

iX is an Autonomous GPT-4 Agent Platform designed to let users drive agents through chat to investigate, plan, and complete tasks.

Purpose

• An agent platform for building, testing, and running autonomous agents.
• Supports task types such as code generation, data extraction, data augmentation, and research assistant.
• Custom agents and chains can be built to assist with various codable tasks.

Main capabilities

• No-code agent editor: Represents an agent's cognitive logic by dragging nodes and connecting them into a graph; built-in chat makes it easy to quickly test and debug.
• Multi-agent chat interface: Create your own team of agents and interact with multiple agents through a single interface. By default, it includes an IX moderator agent responsible for task delegation; you can also @ mention a specific agent to complete a task. The smart input bar supports autocomplete for agent @ mentions and files and data {artifacts} produced by tasks.
• Message queue-driven agent workers: The agent runtime backend runs in Docker and is triggered by a Celery message queue, supporting horizontal scaling to run a large number of agents in parallel.
• Component configuration layer: Maps LangChain components to configuration graphs, used to dynamically render nodes and forms, etc. in the no-code editor.
• Multi-model support (experimental): Google PaLM, Anthropic, Llama.

How to use

Quick start: Use the agent-ix CLI to launch a preconfigured Docker cluster via docker-compose, automatically download the required images, and start the application cluster. The scale command can be used to scale the number of agent workers (limited by available memory and CPU). Clients can specify a specific version, including unstable development images built from the master branch.

Basic workflow:

1. Set up the server and access http://0.0.0.0:8000 , and the system will automatically create a new chat containing the default agents.
2. Enter a request, and the IX moderator will delegate the task to the most suitable agent; or @ mention a specific agent.
3. Custom agents can be added or removed as needed to handle tasks.

Creating custom agents and chains:

• Build using the chain editor or the Python API.
• Navigate to the chain editor, click the root connector to open component search, drag components such as agents, chains, and tools into the editor, connect the components to form a chain, and open the test chat to try it out. Documentation for the chain Python API can be found in the project repository.

Development environment setup (see the project repository for details):

• Prerequisites: Install git and make (Mac: brew install git make; Linux/WSL: apt install git make).
• Clone the repository, set the configuration in .env, and build and initialize the IX cluster (images are built automatically if necessary; set NO_IMAGE_BUILD=1 to skip rebuilding).
• Use make commands to manage the development environment, such as make up/make cluster to start the application (http://0.0.0.0:8000），make server to view web and application container logs, make worker to view agent worker logs, make image to build Docker images, make frontend to rebuild the frontend, make migrate to run Django database migrations, etc.
• Use the ./manage.py dump_agent -a alias command to dump agents and chains (including the component graph).

Windows users can first run wsl --install to install/activate WSL.

Use cases

• Need autonomous agents to complete tasks such as research, code generation, data extraction, and enrichment.
• Want to handle complex tasks through multi-agent team collaboration.
• Need a no-code way to quickly build and debug agent logic.
• Need to horizontally scale agent workers to process large numbers of tasks in parallel.

Limitations

• Google PaLM, Anthropic, and Llama support is marked as experimental.
• The number of workers is limited by available memory and CPU capacity.
• Regarding fees, pricing, and specific platform compatibility, etc., no information is provided in the supplied materials.