Skip to content
View original post on X: meng shao
Published
Pick72/100

MiniMax open-sources OpenAgentCore, a self-hosted OpenAI Agents API

AIMiniMax has open-sourced OpenAgentCore, a self-hosted OpenAI Agents API. Developers can point the official OpenAI SDK at its Core /v1 endpoint, choose harnesses such as Codex, Claude Code, or MiniMax Code per session, and swap models and sandboxes through defined protocols.

AIWhy it matters.

The post shows how a self-hosted agent API lets one OpenAI-compatible interface run several agent harnesses and swappable models, sandboxes, and storage behind it.

meng shaoVerified on X
@shao__meng
View on X

MiniMax has open-sourced OpenAgentCore, a self-hosted OpenAI Agents API: one Core, multiple Harnesses. With the official SDK you only change OPENAI_BASE_URL, and within a Session you can natively run Codex / Claude Code / MiniMax Code; the model and sandbox are swappable under the protocol.

Open source repo:
https://github.com/MiniMax-AI/OpenAgentCore

Main problems it aims to solve
1. Same API: use the official OpenAI SDK or raw HTTP, pointed at your Core /v1.
2. Your Agent + your model: each Session selects a harness (e.g. extra_body.agent.x_agents_core.harness = "codex"), and the model goes through the provider you configured.
3. Compute can sleep, workspace can persist: microsandbox releases compute when idle, and files persist separately, restorable on compatible nodes; E2B can pause in place; Docker and local Linux/macOS/Windows are also supported.
4. Replaceable boundaries: Sandbox Provider, workspace filesystem, Harness, and model provider all go through defined protocols (see AGENTS.md).

Project architecture
· Core + PG: authentication, frozen configuration, Turn scheduling, cancellation/interaction, persisted execution facts
· Sandbox Provider / Node: request and reclaim compute; nodes run Docker or microsandbox
· Workspace FS: persistent environment storage (separate from compute lifecycle)
· Runtime: prepares the environment and capability snapshot, runs the Session Executor, reports events and receipts
· Harness adapter: calls upstream SDKs/protocols and translates events; the native agent loop and context belong to the Harness
· Model provider: Responses / Anthropic / chat_completions and other protocols
· Web: operations console (goes through /core/v1)

AITranslated from Chinese
SaladDay@saladday
[1/n] Previously, people called APIs Tokens as a Service; now it's gradually becoming Agents as a Service. But building this layer has a very practical problem: models and harnesses basically iterate every few months, so if you tie your business directly to a specific implementation, swapping things later becomes very painful. So the underlying Agent Infra needs to keep its external interface as stable as possible, while internally being able to swap in the most frontier components at any time. This is why we built OpenAgentCore, an open-source Agent Infra: externally it provides only one standard API, internally the harness, model, sandbox, and Filesystem are all protocolized, so they can be swapped freely without being locked to any one vendor. https://github.com/MiniMax-AI/OpenAgentCore
AITranslation · show original

[1/n] 以前大家调 API 是 Tokens as a Service，现在慢慢变成了 Agents as a Service。 但做这层存在一个很现实的问题：现在模型和harness基本几个月就迭代一轮，如果把业务直接绑死在某套实现上，后面换起来会非常痛苦。所以底层的 Agent Infra 尽可能需要做到对外的接口足够稳定，但是对内得能够随时换上最 frontier 的组件。 这就是我们做 OpenAgentCore 的初衷，一个开源的 Agent Infra：对外只提供一套标准的 API，对内不管是 harness、model、sandbox 还是 Filesystem 全部协议化，都可以随意更换，不会被绑在某一家上。 https://github.com/MiniMax-AI/OpenAgentCore

View quoted post on X
Source details

meng shao · x.comPublished · added here

Original title: MiniMax 开源了自托管的 OpenAI Agents API「OpenAgentCore」：一套 Core，多种 Harness；官方 SDK 只改 OPENAI_BASE_URL，Session 里原生跑 Codex / Claude Code / MiniMax ...

Topics

MiniMaxAI agentsOpen-source ecosystemDeployment & engineeringMCP & tool useChina AIProduct updates