Flagline scores text against a moderation policy you control and returns a verdict: `allow`, `flag`, or `block`. A request sends the text, a saved preset or an inline policy, and a decision level. Every completed decision returns category scores, the pinned model version, and an audit record.

Flagline recommends a verdict. Your application decides how to enforce it.

<Cards>
<Card title="Get started" href="/docs/getting-started">
Create a workspace, publish a preset, make a key, and send a first request.
</Card>

<Card title="API reference" href="/docs/api">
Request and response shapes, limits, idempotency, and error codes.
</Card>

<Card title="Presets" href="/docs/presets">
How categories, thresholds, severity, and versions fit together.
</Card>

<Card title="Launch status" href="/docs/status">
What is implemented today, and what is deliberately not shipping yet.
</Card>
</Cards>

## The pieces [#the-pieces]

- **Workspace.** Owns presets, API keys, decisions, and usage.
- **Preset.** A named policy made of categories. Each category has a description, optional examples, a flag threshold, a block threshold, and a severity.
- **Version.** A published preset is immutable. Editing produces a browser-tab draft; publishing creates a new version, and the active version changes only when you publish or roll back.
- **Decision level.** `light`, `standard`, or `heavy` shifts the category thresholds and selects how inference runs. The level does not create a second set of thresholds.
- **API key.** A tenant-scoped credential with allowed levels, rate and concurrency limits, and an optional daily spend ceiling.

## Start here [#start-here]

<Steps>
<Step title="Create or open a workspace">
A workspace has to be mapped to a Flagline tenant before presets, keys, and usage can load. Create one from the Workspaces page if you do not have one yet.
</Step>

<Step title="Choose a preset">
System presets are read-only starter templates. Fork one into your workspace to change it, or create a team preset from scratch. Publishing requires a test of that exact draft at its default level.
</Step>

<Step title="Make an API key">
The key needs at least one allowed level. Copy the secret when it is shown; the dashboard does not display it again.
</Step>

<Step title="Call the moderation endpoint">
Send `POST /v1/moderations` with `Authorization: Bearer <key>` and either `preset_id` or an inline `preset`.
</Step>
</Steps>

<Callout type="warning" title="Current launch status">
The v1 HTTP surface is implemented, but it has not been live-tested against Baseten, prepaid
billing, shared caching, or production load. Treat it as not production-ready and read [Launch
status](/docs/status) before relying on it.
</Callout>