omnicoreagent
An agent runtime for Python. Give your agent real work. Keep control: every action checked before it runs, every run survives a crash without silently redoing a
What is omnicoreagent?
An agent runtime for Python. Give your agent real work. Keep control: every action checked before it runs, every run survives a crash without silently redoing anything, every step on the record.
Overview of omnicoreagent
What it is
OmniCoreAgent is an agent runtime for Python. It includes the harness, loop, tools, context, workspace and sub-agents around the model, and adds what running an agent for real needs: a policy on every action, a sandbox as the boundary for its commands, runs that survive a crash, budgets, and a record of every step. It can be used from Python, the command line, or as an HTTP server.
Key features
- Harness: the loop around the model, including tools (yours and MCP), context and memory, a workspace, code mode, skills, sub-agents and worker profiles - Control: a policy (allow, ask a person, or refuse) decided before every action, approvals, budgets, a sandbox that is the boundary for commands - Durability: pause and resume, survive a crash without redoing work, take over from a dead process - Record: one trace per run, readable start to finish, private by default, exported where you want it (OTLP, LangSmith, Opik, JSONL)
- Code mode: a
run_codetool that lets the model write short Python programs, run in Monty, with every call governed and traced - Governance: a policy over every capability (tools, MCP servers, sandbox, network, delegation, background runs);askpauses the run for a person; shell commands can be matched by prefix rules - Execution: anexecutetool whose commands run in a sandbox (Docker, E2B, Modal, Daytona, Vercel, your own, orlocal) with no network unless the policy allows it
Use cases
- From a first script to a governed background worker, and onto a benchmark - Headless runs for CI and scripts: one instruction, a terminal state, an exit code, and the result and trajectory on disk - Running on Harbor (the framework Terminal-Bench runs on): the agent is installed into each task's container, and every trial reports reward, cost, tokens and a trajectory - Production needs: tools, code mode, governance, execution, and the record of every step
Getting started
bash pip install omnicoreagent # Python 3.12–3.14; check with python --version export LLM_API_KEY=your_api_key # the key for the provider in model_config python import asyncio from omnicoreagent import OmniCoreAgent, ToolRegistry tools = ToolRegistry() @tools.register_tool('lookup_order') def lookup_order(order_id: str) -> dict: return {'order_id': order_id, 'status': 'shipped', 'carrier': 'DHL'} agent = OmniCoreAgent( name='support', system_instruction='You answer questions about orders, using the tools. Answer in plain text.', model_config={'provider': 'openai', 'model': 'gpt-5.6-terra'}, local_tools=tools, ) async def main(): result = await agent.run('Where is order 1042?', session_id='customer-7') print(result['response']) await agent.cleanup() asyncio.run(main())
License and pricing
MIT license. Open source.
Based on the project's public README and repository metadata.