AI
Your console has an AI panel at /ai with two distinct experiences behind one UI:
- AI Chat — talk to a language model. Pick a provider and model, type, get a streamed reply. Sessions persist per organization.
- Org Agent — talk to your infrastructure. A per-organization agent answers questions like "what's deployed right now?" or "why did the last deployment fail?" by querying the live platform API through curated read-only tools.
The rule of thumb: chat answers from the model's head; the agent answers from your cluster.
Which AI feature do I want?
There are five AI workflows. They share the word "AI" but do different jobs:
| You want to… | Use | Where | Details |
|---|---|---|---|
| Ask a model a question or draft text | AI Chat | the /ai panel — needs a provider first: Platform AI or your own | AI Chat |
| Ask about your infrastructure: what's deployed, why a deploy failed | Org Agent | the /ai panel; admins provision it under Settings → AI → Org Agent | The Org Agent |
| Use your own AI tool (Claude Code, Cursor) against your organization | MCP token | Settings → AI → Connect your AI assistant | Connect your AI assistant |
| Let an app you deployed call AI models | AI Model Access | Create Service → AI Model Access | AI for your own applications, and Prompt caching to cut its cost |
| Scaffold a new service from a description | AI Assistant template generator | Create Service → AI Assistant | Generate a service |
Two different things are called AI Assistant: the /ai panel this page is named after, and the template generator in Create Service. They are unrelated.
AI Chat, the Org Agent, MCP tokens and AI Model Access need the AI assistant feature on your plan; the models you can use depend on your plan's AI tier (see Usage and spend). The template generator only needs an AI provider configured under Settings → AI. Tiers, providers and spend caps are configured by whoever runs the platform. See AI for operators and Manage plans.
Generate a service with AI
Create Service → AI Assistant, in a project environment, is a template generator, not the /ai panel. It works in three steps:
- Describe your needs: the kind of service, its purpose, and any features you want.
- Choose a variant: pick one of your organization's AI providers, and it proposes variants, each a Docker Compose file with its environment variables.
- Review and finalize: check the Docker Compose file, environment variables and configuration files before you finish.
It only helps you author the service. Nothing it creates calls AI at runtime; for that, use AI Model Access. It runs on the provider you picked, so Platform AI usage counts against your plan like the chat panel, and your own provider bills you directly.
Usage and spend
Model usage through Platform AI and the Org Agent is metered per organization at the platform's model gateway. Your plan's monthly spend cap is enforced at the gateway: when the cap is reached, requests are declined until the period rolls over or an operator raises the cap (raising it takes effect within minutes, with no re-setup). BYO provider usage is never metered or capped by the platform — it's between you and your vendor.