Sources are the "raw material" of your pipeline: the system checks them regularly, collects new materials and files them into the project. From there, you (or Autopilot) decide what becomes posts.
Source types
- Telegram channel. Collects new posts from public channels. Cost β 2 tokens per collection run.
- Website. The system tracks new articles appearing on a site and pulls in their text and images (5 tokens per run). Works even where there is no RSS.
- RSS feed. The classic option for news sites and blogs (5 tokens per run).
- YouTube channel. New videos are automatically transcribed into text (2 tokens per minute of video) β an hour-long video yields a full article to build on.
- Recurring page. A single URL with changing content β exchange rates, weather forecasts, event listings. The system fetches the latest version on a schedule (3 tokens).
- AI generator. A source with no external data: you write a prompt ("come up with an unusual space fact and explain it") and the system creates materials from scratch.
- Aggregator. Builds digests and roundups from materials already accumulated in the project, following your rules β handy for weekly recaps.
There is also one-off import: you can manually pull in a single YouTube video (transcribed into an article) or a single page by URL.
Schedule and runs
Each source has its own check schedule: from "several times a day" to "once a week". You can also start a collection manually with a button. The run journal shows what was collected and when, with real-time progress.
What happens to collected material
Every material lands in the project with "draft" status β it's not a post yet, it's raw material. From there, three paths:
- Manually: browse the list, accept what's interesting, reject the rest.
- Auto-accept: the source marks everything it collects as accepted.
- Auto-analysis: the AI scores each material (1 token) and accepts only those that clear your quality/relevance threshold β the sweet spot for busy sources.
Order and cleanliness
To keep the project from drowning in raw material, automatic cleanup runs: old unaccepted drafts are rejected, rejected items are archived, and the archive is eventually deleted (retention periods are configured at the project level). Files are deleted only at the final step β until then, a material can be restored.
A real example
The niche channels "Azbuka | Fruit Garden" and "Azbuka | Flowers" follow the typical "a few sources + AI generator" pattern: specialist websites supply the news, the generator supplies evergreen tips, and the schedule keeps a steady publishing rhythm. In total, more than 40 active sources are running in the system right now.
Next up: how an accepted material becomes a post β "Working with Content".