Repolode
Submit
omnirexflora-labs

omnirexflora-labs/omnicoreagent

A Python agent runtime that wraps an LLM with tools, memory, a policy on every action, a sandbox for commands, crash-safe runs and a full trace of each step.

View on GitHubWebsite
Stars
252
Forks
57
Open issues
0
Last commit
13 days ago

What is it?

OmniCoreAgent is a runtime for building LLM agents in Python. It covers the parts around the model that an application needs: a tool loop with your own Python functions and MCP servers, session memory, context management, a workspace for files, and sub-agents. It can be used from Python code, from a command line, or as an HTTP server.

The runtime adds control for agents that take real actions. A policy with allow, ask and deny rules applies to tools, MCP servers, the sandbox, the network and delegation, and `ask` pauses a run for a person's approval. Commands run in a sandbox provider such as Docker, E2B or Modal, and budgets limit spending in dollars, tokens or calls. Runs keep a checkpoint so that a run whose process died can resume without silently repeating interrupted tool calls.

Each run produces a trace of model calls, tool calls, approvals and budget decisions, which can be exported to OTLP, LangSmith, Opik or JSONL. Outcomes can be attached to a run later, and finished runs can be returned as training records.

Key features

  • 01Tool loop supporting Python functions and MCP servers over stdio, SSE and streamable HTTP.
  • 02Policy with allow, ask and deny rules covering tools, sandbox, network and delegation.
  • 03Command execution in sandboxes such as Docker, E2B, Modal and Daytona.
  • 04Budgets in dollars, tokens or calls, per request, session or day.
  • 05Crash-safe runs that resume from checkpoints without repeating interrupted calls.
  • 06Per-run traces exportable to OTLP, LangSmith, Opik or JSONL.

Use cases

  • Building a customer-support agent that calls internal APIs and needs human approval for refunds.
  • Running a coding agent in CI that fixes failing tests inside a sandbox with budget limits.
  • Serving agents over HTTP with REST and SSE endpoints for background tasks.
  • Auditing agent behavior by reading full traces of model and tool calls.

Quick start

  1. 1

    Install the package with pip

    pip install omnicoreagent
  2. 2

    Set your LLM provider API key

    export LLM_API_KEY=your_api_key
  3. 3

    Save the README quick start code as a Python file and run it

Things to know

  • Requires Python 3.12 to 3.14; installation stops on Python 3.10 or 3.11.
  • Needs an API key from a supported LLM provider to run agents.

Topics

#agent#llm#python#mcp#sandbox#agent-runtime#observability#agent-harness#ai-agents#background-tasks

FAQ

What is omnicoreagent?
A Python agent runtime that wraps an LLM with tools, memory, a policy on every action, a sandbox for commands, crash-safe runs and a full trace of each step. OmniCoreAgent is a runtime for building LLM agents in Python. It covers the parts around the model that an application needs: a tool loop with your own Python functions and MCP servers, session memory, context management, a workspace for files, and sub-agents. It can be used from Python code, from a command line, or as an HTTP server.
Is omnicoreagent free to use? What license does it have?
omnicoreagent is open source under the MIT license. Check the license terms before using it in commercial products.
Is omnicoreagent still maintained?
Actively maintained. The last commit to omnicoreagent was on Oct 9, 2026.
What can I use omnicoreagent for?
Building a customer-support agent that calls internal APIs and needs human approval for refunds. Running a coding agent in CI that fixes failing tests inside a sandbox with budget limits. Serving agents over HTTP with REST and SSE endpoints for background tasks. Auditing agent behavior by reading full traces of model and tool calls.
How do I install omnicoreagent?
Install the package with pip: pip install omnicoreagent Set your LLM provider API key: export LLM_API_KEY=your_api_key Save the README quick start code as a Python file and run it
Where can I get omnicoreagent?
The source code and installation instructions are on GitHub: https://github.com/omnirexflora-labs/omnicoreagent

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects