Per-hour agent billing, Control Tower, and agents that see into your files
A big week: agent billing moves to a new model, Control Tower separates obligations from awareness, and agents proactively act on attachments you drop in a space.
Agent billing by the hour
Agent runtime now bills by the hour, not by the token. Three tiers, three rates:
- Frontier — $10/hr. The highest-capability models for complex reasoning, writing, and research.
- Balanced — $4/hr. Strong all-around models for day-to-day tasks.
- Cost-Effective — $2/hr. Fast, lightweight models for high-volume, lower-complexity work.
Do and Know in Control Tower
Control Tower now makes the product distinction explicit. Do holds work that needs an outcome and is not cleared by reading; Know assembles grounded awareness into an edition you acknowledge as a whole.
Three details make the model reliable:
- Recipient-relative ranking. The same message can matter directly to you and have different relevance through several Library topics without those assessments overwriting each other.
- Generation-scoped obligations. A retraction closes only the current ask; a genuinely new ask on the same topic can appear again.
- Exact unread discharge. Know clears from positioned surface progress, not timestamps or lifetime group totals.
Agents that respond to file drops
Drop a PDF, Word doc, spreadsheet, or text file into any space and the right agent activates. Monad now reads the file name, type, and a preview of its contents (this was fun to build!) before deciding which agent(s) should respond — so it knows what you dropped without you having to explain it.
Improvements
- Delete canvas database views. Right-click any view tab inside a canvas database to delete it. A confirmation dialog shows the view name before anything is removed. The last remaining view on a database can't be deleted — at least one always stays.
- Onboarding confirms what Monad found about your company first. Before the setup wizard begins, you see a read-only summary of what Monad researched and can confirm it looks right — so your workspace is set up around the right company from the start.
- Plan editor requires confirmation before deleting. Remove buttons in the workspace plan editor are always visible (no longer hover-only), and all removals — plan rows and pending invites — need confirmation before they take effect.
- Company research no longer delays attention delivery. Long-running research tasks now run separately, so Control Tower, Herald, and search stay fast even while Monad is looking something up.
Fixes
- Agents now retry when a response is cut off mid-stream. Truncated AI responses previously failed silently. Monad detects the truncation and retries automatically.
- Space names are cleaned up automatically. If Monad proposes a space called
#general, it's saved asgeneral— the leading#is stripped. - Onboarding won't proceed with an incomplete plan. A built-in check requires the proposed workspace plan to be confirmed complete before setup can continue.