Guides
Connect an OpenAI-compatible service (Kimi, GLM, Ollama…)
Any service that speaks the "OpenAI format" works: Moonshot (Kimi), Qwen, Zhipu GLM, and Ollama running on this computer. Agents Workshop requires a model that can see images, so pick a vision / VL model.
OpenAI itself, Anthropic and Alibaba Cloud Model Studio each have their own card (which also explains subscription sign-in and plans) — no need to use this one for them.
What you need
- An account with that provider and an API key from its console;
- The "base URL" from its docs (usually ends in
/v1).
Consoles of the common ones:
Steps
- Create an API key in the provider's console.
- Find the base URL in its docs (usually ending in /v1).
- In Agents Workshop, click "Add API key" on the "OpenAI compatible (custom)" card. The presets at the top of the form (Moonshot / Kimi, Qwen, Zhipu GLM, local Ollama) fill in the URL and a vision-capable model in one click.
- Fill in URL, model name and key; click "Fetch model list" if you don't know the model name.
- Save, then click "Test" — three ticks: reachable / text replies / reads images.
Preset addresses
| Preset | Base URL | Default model |
|---|---|---|
| Moonshot / Kimi | https://api.moonshot.cn/v1 | moonshot-v1-8k-vision-preview |
| Qwen | https://dashscope.aliyuncs.com/compatible-mode/v1 | qwen-vl-plus |
| Zhipu GLM | https://open.bigmodel.cn/api/paas/v4 | glm-4v-plus |
| Local Ollama | http://127.0.0.1:11434/v1 | llama3.2-vision |
Local Ollama needs no key, but Ollama must be installed on this computer with that model pulled.
FAQ
- The test says it can't read images: switch to a vision-capable model and test again.
- Prices: models in the built-in price list are filled in automatically; others can be typed by hand, and prices marked "manual" are never overwritten by a refresh.
- The API key only goes into this computer's encrypted store: never through AI, never into logs.