OmniCoreAgent 🤖 Agent Open source

What it does

OmniCoreAgent is an open-source Python framework for building AI agents that hold together in production. It wraps a language model with the parts an application needs around the model loop: a prompt contract, local Python tools, MCP server tools, memory, context management, a workspace for files, guardrails, and events. It can run several independent tool calls in one batch and turn the results into a single structured observation, and it detects when an agent is stuck in a loop using hashed signatures of tool calls rather than a plain step count. Heavier pieces such as durable background tasks and an HTTP/SSE serving layer are optional, so a small agent stays small.

What is inside

• The agent harness: model loop, tool calling, structured observations, memory, context control and guardrails
• Tools: local Python functions and MCP server tools through one runtime surface
• Subagents and workflow orchestration for multi-step tasks
• Background tasks with a durable task store, run history, retries and cancellation
• OmniServe: REST and SSE APIs with readiness, auth, rate limits and metrics
• Optional backends for Redis, MongoDB, SQL and S3-compatible storage, installed as extras

Works with

Any model provider you configure. It is a library you build applications on, not an extension for a specific coding-agent app.

How to install

pip install omnicoreagent

Install extras such as pip install "omnicoreagent[serve]" only when you need them.

Maintenance and safety

The project is MIT-licensed and actively developed. It routes tool output through an observation pipeline that can offload large payloads and apply guardrails, which helps contain prompt-injection content from tools. You supply your own model API key, and any MCP servers you connect carry their own access and risks. The cookbook is extensive but the API is large, so expect a learning curve.

Who should use it

It suits Python developers building agents that need parallel tools, background work and a serving boundary, not just a demo tool loop. If you only want to add skills to an existing coding agent, a lighter tool is a better fit.