HopBase DocsWorkBuddy
Use HopBase as the model backend behind WorkBuddy through its OpenAI-compatible provider settings.
WorkBuddy can call an OpenAI-compatible endpoint as its underlying model backend, so HopBase can serve the models behind it. Verified in production: WorkBuddy reaches POST /v1/chat/completions and streams normally.
Connection settings
| Setting | Value |
|---|---|
| Protocol / provider type | OpenAI (Chat Completions) |
| API address / Base URL | https://api.hop-base.com/v1 |
| API key | A plan key for a chat model group |
| Model | The exact ID returned by GET /v1/models for that key |
Setup
Pick a chat-capable plan key
Sign in to the console and use a key whose group serves chat models — Claude, GPT / Codex, GLM, Kimi, DeepSeek, Qwen, or Grok.
Read the exact model ID
curl https://api.hop-base.com/v1/models \
-H "Authorization: Bearer sk-your-key"Copy an ID from the response verbatim.
Fill in the provider
In WorkBuddy's model provider settings, choose the OpenAI-compatible type, then enter the base URL, the key, and the model ID from the steps above.
Send a test message
Ask for a one-word reply. A successful round trip confirms the key, the group, the protocol, and the model ID all line up.
Two mistakes that come up first
A media group key cannot serve chat
Pointing WorkBuddy at a key from an image or video group — Kling, Seedance, MiniMax, Wan / HappyHorse — returns route_not_found. Those groups serve task-based media endpoints, not chat/completions. Use a chat model group instead.
Short model names are rejected
A family name such as glm returns 404 model_not_found. Model IDs are exact and dated or versioned, for example glm-5.3 rather than glm. Always copy the ID from GET /v1/models for the key you configured.
Claude models
Claude uses the Anthropic protocol, not Chat Completions. If WorkBuddy only offers an OpenAI-compatible provider type, point it at a chat group that serves OpenAI-protocol models. See Third-party clients for the protocol-per-family table.
Long system prompts
WorkBuddy sends a large instruction context, so a single turn can carry tens of thousands of input tokens. Pick a model whose context window has room for it, and be aware that some families change pricing tier once input crosses a threshold — see the family's own page under Model integration.
Related
- Third-party clients — the same settings for Cherry Studio, NextChat, and LobeChat
- Available models — what each plan group serves