Intents

Reusable analysis recipes that turn raw audio into structured insights and downstream actions.

An intent captures what to extract from a recording and where to send the result. It's the bridge between a transcript and a useful outcome — a GitHub issue, a Slack post, a Notion page, an Anki card, or just a downloadable artifact.

Anatomy of an Intent

Every intent has:

• A name and description (used in the channel attach picker and dashboards)
• An analysis prompt that the LLM runs over the transcript
• A provider target — the destination for the result
• A status: active, paused, draft, or error

Supported provider targets include:

Creating an Intent

From Intents → New, configure the prompt and target. The new-intent flow validates that:

1. The selected provider has at least one active connection (see Integrations)
2. The prompt is non-empty and within length limits
3. Required provider-specific fields are filled (e.g. GitHub repo, Slack channel)

Drafts can be saved before everything is wired up.

Running an Intent

Intents run in three ways:

1. Automatically when an audio finishes processing on a channel that has the intent attached
2. Manually by selecting an intent on the audio detail page
3. Via the API by triggering the underlying Trigger.dev task

Each run produces an Insight record that tracks status, output, and integration delivery logs.

Filtering and Status

The intents index supports filtering by provider and shows a per-intent status indicator:

• Active — Will fire on new recordings
• Paused — Configured but skipped on new audios
• Draft — Incomplete; cannot run yet
• Error — The last run failed; check the linked insight for details

Best Practices

• Keep prompts focused — one intent per outcome (action items, summary, FAQ entry)
• Pause rather than delete intents when iterating, so existing insights keep their lineage
• Use the download provider for prompts you want to test before wiring an integration