The AI assistant
Use your own AI key
Connect OpenAI, DeepSeek, OpenRouter or any OpenAI-compatible endpoint and stop spending chat credit.
If you already pay an AI provider, you can point DevOps Agent at your own account. Chats then bill to your provider account, spend no DevOps Agent credit, and the meter reads "Your key".
This works with anything that speaks the OpenAI API format: OpenAI itself, DeepSeek, OpenRouter, Groq, Together, a Gemini-compatible endpoint, or your own hosted model.
Adding one
Settings → AI model → Add your model.
| Field | What to enter | Example |
|---|---|---|
| Base URL | The provider's API address | https://api.openai.com/v1 or https://api.deepseek.com/v1 |
| Model name | The exact model id, as the provider spells it | gpt-4.1, deepseek-chat |
| API key | Your key from the provider | sk-… |
Press save. We test it with one short message before storing it, so you find out immediately if the URL, the model name or the key is wrong rather than in the middle of an incident.
Requirements
- The base URL must be
https://and a public address. Private and internal addresses are refused, because our servers have to be able to reach it. - The model must support tool calling. The assistant works by calling tools — running commands, reading files. A model that cannot call tools will chat but will not do anything. Most current models from the major providers can.
Where to get a key
| Provider | Where |
|---|---|
| OpenAI | platform.openai.com → API keys |
| DeepSeek | platform.deepseek.com → API keys |
| OpenRouter | openrouter.ai → Keys (one key, many models) |
You are billed by them, at their prices, and the usage appears on their dashboard.
Scope and sharing
A model you add belongs to your workspace: it is available in every project in it, and every member of your team can select it. The key itself is encrypted and is never shown again after saving — you can replace it, but not read it back.
Managing them
From the same screen you can edit, delete, and set one as your default for new chats.
Deleting a model does not break existing chats; they fall back to your default.
Which is better?
Our hosted models are tuned and priced for this product, need no setup and work out of the box.
Your own key makes sense if you already have provider credit, want to use a specific model, or are a heavy user who would rather pay the provider directly.
Either way the safety switch, the approval cards and the activity log work identically — none of that depends on which model is answering.
Permissions
Adding, editing and deleting models needs the Owner or Admin role. Everyone can choose among the models available and set their own default. See Roles.