AI Chat
Talk to a language model from the /ai panel. Sessions persist per organization.
Providers
The chat panel works with provider rows configured under your organization's AI settings. Two kinds coexist:
| Provider | Who sets it up | Whose key & bill | Caps |
|---|---|---|---|
| Platform AI | An owner or admin, once — see below | The platform's, on your plan's AI allowance | Your plan's tier and monthly spend cap |
| BYO (bring your own) | You, in AI settings | Yours (your OpenAI/Anthropic/compatible key) | None from the platform |
Turn on Platform AI for chat
Until somebody does this, AI Chat says "No AI provider is configured for this organization" — even when your plan includes the AI assistant. The plan entitles you to the models; this is the switch that puts them in the chat panel.
An owner or admin does it once:
- Open Settings → AI.
- In Platform AI for chat, click Enable Platform AI.
- Open the
/aipanel — Platform AI is now selectable as a provider, for everyone in the organization.
Then:
- Nobody needs an API key of their own. Usage counts against your plan's AI allowance and stops at your plan's monthly cap.
- Which models you get follows your plan's AI tier: the platform's built-in models on the free tier, premium models on paid tiers where the operator has configured them.
- The row is managed — it shows up in your provider list but can't be edited or deleted like a BYO row. Remove it from the same card with Remove from chat.
- Provisioning the Org Agent turns this on too, because the agent runs on the same allowance. If you already provisioned the agent, Platform AI is already in your chat panel and the card will say so.
BYO providers are ordinary rows you create with your own vendor key. Use them when you want a specific model, no platform caps, or your own billing relationship. Your key is encrypted at rest and never shown back once saved. BYO rows and Platform AI coexist — turning one on never disturbs the other, and each chat session records the provider it was started with.
How the assistant answers
AI Chat answers from the model's own knowledge. It cannot see your projects, applications, deployments or logs — that is the Org Agent, which reads them through curated read-only tools. Ask chat about the state of your infrastructure and it will tell you to use the agent rather than guess.
Answers are kept brief by default — a short paragraph unless you ask for more, with commands and configuration in code blocks you can copy.
To change that for a whole conversation, choose Answer length when you start the chat:
| Choice | What you get |
|---|---|
| Brief | One to three sentences — the answer, without background or alternatives |
| Normal (default) | A short paragraph |
| Detailed | The reasoning, the relevant options and their trade-offs |
Like the model, it is fixed for the life of the chat — start a new one to change it. Within any setting you can still ask for more or less in a message ("explain step by step"); the choice sets the default, it does not cap the answer.
To copy an answer, hover it and use the copy button beside it. It copies what the model wrote, without the surrounding page.
Which model a chat starts on
Each chat picks its model when you start it, and keeps it for the life of that conversation — changing model means starting a new chat.
New chats start on your plan's default: the built-in free model on the free tier, and on a paid tier the lowest-cost premium model your operator offers. The picker lists every model your plan can reach, so a paid organization can still start a chat on the free model when the question doesn't need more — useful for keeping spend down.
The Org Agent works differently, because it runs one model for every question it answers — there is no per-message choice. Pick it in Settings → AI → Org Agent:
- Agent model chooses the lane: your plan's models, or one of your own providers.
- On your plan's lane, Model then chooses which one. The list marks your plan's default, the lowest cost and highest cost options, and the free model as costing nothing against your allowance. Changing it takes effect when you Re-provision.
A cheaper model costs less per answer, and usage counts against the same allowance either way — worth knowing, since an agent answering a question may make several model calls to gather what it needs. If you never choose, the agent stays on your plan's default.