The AI assistant
Models and credit
Which AI model to pick, what a chat costs, and what happens when credit runs out.
The assistant costs real money to run — the panel does not. So the chat has a visible credit meter and you choose which model answers.
The hosted models
They appear under DevOps Agent in the model picker and spend your chat credit. Which ones you get depends on your plan.
| Model | Tagline | Plans |
|---|---|---|
| luna | Cheap and good | Every plan — the only one on Free |
| terra (default on paid plans) | Moderate but good day to day | Pro ($19) and above |
Rough cost per answer: luna a fraction of a cent, terra around three cents. A typical investigation is several answers.
On Free, terra shows in the picker with a lock and a link to Settings → Subscription.
The credit meter
The bar near the message box shows how much of your credit you have used. Open it for a breakdown: the cost of the last answer, token counts, and the total for the chat.
New Free accounts get $2 of credit — several hundred answers on luna.
When credit runs out
You see: "Chat credit is used up. Upgrade your plan in Settings → Subscription, or switch to one of your own models."
Nothing else stops. Every screen, every server tab, the terminal, uptime — all of it keeps working. Only the assistant pauses. Your options:
- Bring your own AI key — then chats cost you nothing here and the meter reads "Your key".
- Ask your team's owner or admin to raise your limit, if you are on a team.
- Upgrade — Settings → Subscription → Contact us to upgrade, or email support@devops-agent.io.
Teams share one pot
On the Team plan, credit is pooled — everyone spends from the owner's balance. That is deliberate: per-person credit means the junior engineer is blocked at the worst moment while the senior still has plenty.
Making credit go further
- Use the right model. luna is plenty for "what does this error mean".
- Start a new chat per problem. A chat carries its entire history into every reply, so a long-running one gets steadily more expensive.
- Use guided tasks. They reach the answer in fewer turns.
- Be specific. Every round of "which server?" costs money.
Choosing the default for new chats
Settings → AI model has a Default model setting. New chats start with it.
An existing chat keeps whatever model it was started with; the picker at the top of a chat changes that chat only. If a chat's model is deleted, it falls back to your default.
Reasoning effort
Some models expose a reasoning-effort setting in the chat header. It makes the model think longer before answering — better on hard problems, more expensive.
Note: this setting only takes effect on OpenAI-hosted models. On other providers added through your own key, the header may show a level that the provider ignores.