π³ Credits & Limits
AI creditsβ
Building and maintaining a knowledge base consumes AI credits from your workspace's monthly pool, the same pool the assistant's answers draw on. The Owner and Admins can see the pool, set per-member caps and review the activity log on the workspace's AI Usage page. Anything that fetches content from the outside world or runs a model over your knowledge has a cost.
Adding sourcesβ
| What | Charged |
|---|---|
| Website crawl | Per page actually indexed, not per page fetched or seen |
| YouTube | Per video indexed |
| X | Per 10 new posts indexed; posts you already have are free |
| Files, GitHub, chat capture | Not charged |
| A sync that finds nothing new | Nothing, except syncing a single YouTube video again |
| Pages skipped because they were already indexed | Nothing |
| Re-crawl, or Sync now on a website that isn't looking for new content | Per page indexed again, when the site changed |
Before a crawl starts you see how many pages it will fetch, and you're billed only for what reaches your agents, which is normally well under that, because language variants, duplicates and already-indexed pages all drop out along the way.
Maintaining knowledgeβ
Each of these shows in the AI Usage activity log under the name in the first column:
| Activity log entry | What triggers it | Credits |
|---|---|---|
| Knowledge gap answer drafted | Draft an answer on a gap. See Gaps. | 3 |
| Knowledge upkeep proposal drafted | A fix Fren One drafts on its own for a gap with enough evidence. | 2 |
| Knowledge change backtested | Run backtest on a drafted change, once per change, up to 25 replays. | 5 |
| Knowledge editor AI draft | Draft this section, Rewrite this paragraph or Rewrite selection in the editor. | 2 |
| Reply suggested to an admin | A reply suggestion in Telegram or Discord. Asking again for the same message while its suggestion is open, replying with a macro, and grading answers cost nothing. | 3 |
| Faq generation | One completed Find questions from chats run. See Gaps. | 9 |
Questions in the test panel are charged like any other answer.
Your remaining balance lives under AI Usage in Settings: open the workspace switcher and select Workspace settings, then AI Usage.
Limitsβ
These exist so one misconfigured job cannot spend a month's credits in an afternoon.
| Limit | Value |
|---|---|
| Ingestion jobs | 50 a day per channel, or per workspace for sources that belong to no channel |
| Assistant questions | 1,000 a day per channel or workspace |
| Scheduled auto-refresh checks | 500 a month per workspace, shared across sources that keep in sync |
| File upload | 4 MB per file |
| Repositories per GitHub batch | 20 |
| Backtest | Up to 25 past questions and tickets per change |
Hitting the daily ingestion limit shows "Daily ingestion quota exceeded for this channel or workspace". Adding a source and Sync now count toward it; scheduled checks don't. Nothing is lost. The queue resets at midnight UTC, or you can add the remaining sources the next day.
The GitHub cap is per batch rather than per workspace: queue 20 repositories, then run the import again for the rest.
Keeping the cost downβ
- Crawl sections, not whole sites. The path picker exists for this. Machine-generated areas cost credits and make answers worse.
- Keep in sync only what changes. See Auto-refresh.
- Prefer one good source over three overlapping ones. Overlapping sources are deduplicated, so the second copy usually costs a fetch and indexes nothing.
- Check Gaps before adding more. Unanswered questions tell you which gap is worth spending on.