234 ·GitHub
What is this?
OpenAgentCore is an open-source, self-hosted implementation of the OpenAI Agents API that lets you run AI agents on your own infrastructure using pluggable native harnesses like Codex, Claude Code or MiniMax Code.
OpenAgentCore is an open-source, self-hosted implementation of the OpenAI Agents API that lets you run AI agents on your own infrastructure using pluggable native harnesses like Codex, Claude Code or MiniMax Code.Is this a library I import, or a server I run?It is a self-hosted backend service made of a Core and Runtime component; your applications talk to it through the same Agents API protocol as OpenAI, or plain HTTP, rather than importing a library.
It is a self-hosted backend service made of a Core and Runtime component; your applications talk to it through the same Agents API protocol as OpenAI, or plain HTTP, rather than importing a library.How do I get it running?On Linux or macOS run the installer with Docker configured, or use the Windows PowerShell installer; then sign in to Web, the admin console, with the generated Core key, configure the domain/HTTPS, set a default model and issue a Project API key.
On Linux or macOS run the installer with Docker configured, or use the Windows PowerShell installer; then sign in to Web, the admin console, with the generated Core key, configure the domain/HTTPS, set a default model and issue a Project API key.Do I need Docker to install it?On Linux or macOS you need Docker already configured before running the installer; Windows uses a separate PowerShell installer instead.
On Linux or macOS you need Docker already configured before running the installer; Windows uses a separate PowerShell installer instead.Do I need to supply my own model provider and execution capacity?Yes — after installing you still have to set a default model and add execution capacity such as a node, E2B, or your own machine before you can run your first session.
Yes — after installing you still have to set a default model and add execution capacity such as a node, E2B, or your own machine before you can run your first session.Who is this aimed at?It's for teams who want to host and operate AI coding agents themselves rather than depend on OpenAI's hosted Agents API, using their own model providers and infrastructure.
It's for teams who want to host and operate AI coding agents themselves rather than depend on OpenAI's hosted Agents API, using their own model providers and infrastructure.Where do the agents actually execute?Agents run inside a managed sandbox (Docker, microsandbox or E2B), or directly on your own Linux, macOS or Windows machine, depending on how you configure execution capacity.
Agents run inside a managed sandbox (Docker, microsandbox or E2B), or directly on your own Linux, macOS or Windows machine, depending on how you configure execution capacity.Which coding agents does it support?Each Session runs one of three native harnesses — Codex, Claude Code or MiniMax Code — paired with whichever model provider you configure.
Each Session runs one of three native harnesses — Codex, Claude Code or MiniMax Code — paired with whichever model provider you configure.Can I swap out parts of the system?Yes, sandboxes, harnesses and model providers all plug in through defined protocols, so any component can be replaced independently.
Yes, sandboxes, harnesses and model providers all plug in through defined protocols, so any component can be replaced independently.How are long-running sessions tracked?Core keeps durable execution state while the Runtime runs the chosen harness inside the Environment, so sessions can survive restarts and reconnects.
Core keeps durable execution state while the Runtime runs the chosen harness inside the Environment, so sessions can survive restarts and reconnects.Is there an admin interface?Yes, 'Web' is the operator admin console used to sign in with the Core key, configure the domain/HTTPS, and manage models and API keys.
Yes, 'Web' is the operator admin console used to sign in with the Core key, configure the domain/HTTPS, and manage models and API keys.What are similar projects?casbin-gateway, trueforge, harnessrouter, claw-code, cayu, overmind, open-multi-agent, oar
casbin-gateway, trueforge, harnessrouter, claw-code, cayu, overmind, open-multi-agent, oarWhich topics is it in?agent-runtime, durable-execution, openai-api, agent-harness, containers, harness, sandbox, openai-compatible, docker-compose, agents, rest-api, backend
agent-runtime, durable-execution, openai-api, agent-harness, containers, harness, sandbox, openai-compatible, docker-compose, agents, rest-api, backend