# AI

> Anthropic API key, AI feature catalogue, the AI actions switch, and usage.

Source: https://productos.ca/documentation/pages/settings/ai (last reviewed 2026-09-29)

The AI section is where the workspace's AI features get configured. It is gated to **admin** and **owner** roles, members and below see a Locked panel.

## API key

- **Anthropic API key**: paste your `sk-ant-...` key here. It is stored against your user account rather than the workspace (each AI call uses the saved key of the person making it, ahead of any hosted key ProductOS provides), encrypted at rest with AES-256-GCM. After a save the key is loaded back into the field, masked, with a show/hide toggle.
- **Validation**: the key is not checked on save. AI routes resolve it on the first call and answer "No API key configured" if none is stored and no hosted key is provided. A saved key that Anthropic rejects fails the call; it doesn't fall back to the hosted key.
- **Rotation**: paste a new key and save; the old is overwritten in the same write.

## Workspace context

A shortcut to the **Knowledge** page, where you maintain identity, vision, competitors, personas, OKRs, and recent outcomes. Every AI prompt in ProductOS prepends this block as grounding, sparse Knowledge means generic AI output.

## AI features

A read-only catalogue of what unlocks once a key is configured:

- **ProductSense**: per-check data-quality runs across the workspace
- **Copilot Chat**: ask questions about your product data
- **AI Insights**: generate insights for individual entities
- **Release Notes**: auto-generate release notes from a set of features / initiatives

There is no per-workspace model picker. ProductOS picks the model per feature and rolls new Claude versions in centrally; [BYO Anthropic key](https://productos.ca/documentation/guides/ai/byo-anthropic-key) lists which model each feature uses.

## AI actions

**Allow AI to apply changes** is a workspace-wide switch. When it's off, the Copilot can still answer questions and propose changes, but its Apply button and the MCP write tools are blocked for everyone regardless of role. Individual permissions still apply on top. Only the workspace owner can flip it.

## Usage

**AI usage & cost** for the last 7, 30 or 90 days (30 by default): total cost, calls, tokens, spend by source, and a daily chart. Each call is billed to the key it ran on: the caller's saved key, or the hosted key when they have none. The same data is the basis of the cost expectations in [BYO Anthropic key](https://productos.ca/documentation/guides/ai/byo-anthropic-key).

> **Re-run onboarding** lives under **Settings → Workspace** (owner only). It's a workspace lifecycle action (it regenerates vision, competitors, objectives, and a fresh Golden Thread sample), not an AI-key setting.

## Related

- [BYO Anthropic key](https://productos.ca/documentation/guides/ai/byo-anthropic-key)
- [Using ProductOS with Claude Code](https://productos.ca/documentation/guides/ai/claude-code)