Agent Integrations
Agent Integrations
Connect PixMind GPT-5.6 models to OpenCode, Cursor, Trae, Cherry Studio, Cline, and other agent clients.
PixMind exposes an OpenAI Chat Completions-compatible API for language models. Supported agents, IDEs, and desktop clients share one API key, Base URL, and set of model IDs.
Before you start
- Create an API key in the API Platform key dashboard.
- Use
https://aihub-admin.aimix.pro/api-platform/v1as the shared Base URL. - Start with
gpt-5.6-terra, then switch models after the connection works.
Store the API key only in a secure credential store, system keychain, or environment variable. Never commit it to Git.
Supported clients
| Client | Protocol | Current support | Guide |
|---|---|---|---|
| OpenCode | OpenAI Chat Completions | Direct | Configure |
| Cursor | OpenAI Chat Completions | Direct | Configure |
| Trae | OpenAI Chat Completions | Direct | Configure |
| Cherry Studio | OpenAI Chat Completions | Direct | Configure |
| Cline / Roo Code | OpenAI Compatible | Direct | Configure |
| Codex CLI | OpenAI Responses | Gateway required | Read limitation |
| Claude Code | Anthropic Messages | Direct | Configure |
Choose a model
Anthropic
| Model ID | Positioning | Recommended for |
|---|---|---|
claude-fable-5 | Flagship agent | Complex agents, deep reasoning, long workflows |
claude-opus-5 | Advanced reasoning | Architecture, complex planning, difficult debugging |
claude-sonnet-5 | Balanced coding | Daily coding, tool calls, code review |
claude-opus-4-8 | Complex engineering | Large codebases and cross-file refactors |
claude-opus-4-8-cc | Coding agent | Terminal tools and development automation |
OpenAI
| Model ID | Positioning | Recommended for |
|---|---|---|
gpt-5.6-sol | Flagship reasoning | Large codebases, architecture, complex debugging |
gpt-5.6-terra | Balanced default | Daily development, tool calls, mixed agent workloads |
gpt-5.6-luna | Fast and light | Quick answers, small edits, and batch tasks |
| Model ID | Positioning | Recommended for |
|---|---|---|
gemini-3.1-pro-preview | Multimodal reasoning | Complex image and video understanding |
gemini-3.5-flash | Balanced multimodal | Visual Q&A, video understanding, tool calls |
gemini-3.6-flash | Efficient multimodal | Frequent multimodal and batch workloads |
gemini-3.7-flash | Vision and video | Fast content understanding |
gemini-3.5-flash-lite | Light multimodal | Classification, extraction, batch summaries |
Protocol support
PixMind exposes two client protocols:
- OpenAI-compatible clients use
POST /api-platform/v1/chat/completions. - Claude Code uses
POST /api-platform/anthropic/v1/messages, with/v1/messages/count_tokensand/v1/modelsunder the same prefix. - Codex CLI custom providers use OpenAI Responses and still require a Responses-compatible endpoint.
Common troubleshooting
401 Unauthorized: verify the client sendsAuthorization: Bearer <API_KEY>orx-api-key.404 Not Found: the Base URL must end in/api-platform/v1; the client appends/chat/completions.Model not found: use an exact model ID such asgpt-5.6-terra, not a display name.- Streaming errors: verify a non-streaming request first, then enable OpenAI-compatible streaming.