Use Heabsy models in the tools you already run
Heabsy speaks two protocols with one key. Tools built on the OpenAI API connect tohttps://api.heabsy.com/v1; tools built on the Anthropic Messages API, such as Claude Code, connect to https://api.heabsy.com. Each guide below gives the exact settings, the context limits the tool needs to know, and the errors people actually hit.
Coding agents
Chat apps
Frameworks
Not yet: Cursor's Agent mode and Codex CLI need the Responses API, which Heabsy is adding. Aider, LlamaIndex and other OpenAI-compatible clients already work with the base URL above.
What the API supports
| Interface | Endpoint | Supported | Notes |
|---|---|---|---|
| Chat Completions | POST /v1/chat/completions | Yes | Streaming with usage in the last chunk; tool calling on models that support it. |
| Anthropic Messages | POST /v1/messages | Yes | Same key; base URL without /v1. Token counting at /v1/messages/count_tokens. |
| Model list | GET /v1/models | Yes | Returns exactly the models your key may call. |
| Responses API | POST /v1/responses | Not yet | Tools that only speak Responses — Codex CLI, Cursor Agent — cannot use Heabsy today. |
| Embeddings | POST /v1/embeddings | No | Chat and reasoning models only. |
Where requests are processed
The catalog has two tiers behind the same key. Our own models — Qwen3.8 27B, GLM 5.3, MiMo V2.6 Flash and their variants — run on Heabsy hardware in the European Economic Area, and prompts and completions are not stored. Partner-hosted models — Claude, GPT-6, Kimi K3, DeepSeek V4 and others — are served by third-party inference providers without that promise, and each is marked in the model catalog. Prices are per million tokens, with no subscription and no minimum: see pricing. Which models keep nothing, and what is still logged: zero data retention.
FAQ
Is Heabsy an OpenAI-compatible API?+
Yes. Point any OpenAI SDK or tool at https://api.heabsy.com/v1 with a Heabsy key: chat completions, streaming, tool calling and the model list work as in the OpenAI API. The Responses API is not served yet.
Is Heabsy an Anthropic-compatible API?+
Yes. https://api.heabsy.com/v1/messages answers the Anthropic Messages API with the same key, which is how Claude Code and Anthropic SDKs connect without a proxy.
Can I use Cursor with Heabsy?+
Not reliably yet. Cursor routes custom keys through its own servers, and its Agent mode sends requests in the Responses format that we do not serve. Inside VS Code, Cline, Kilo Code and Claude Code work with Heabsy today.
I used Roo Code or Continue — what now?+
Roo Code shut down on 15 May 2026 and Continue became read-only in June 2026. Cline and Kilo Code take the same OpenAI-compatible settings — base URL, key, model ID — and Kilo Code even imports Roo custom modes.