Try it now
A working sandbox. No sign-up, no project. Sample data only.
- One switch per feature, per project: character overviews, connected moments, interview analysis, chat with transcripts, paper cut assist, beat generation and AI voice.
- Noneditorial mode for the reading features. The model may find and quote what was said; it may not summarise a person or take a position on them.
- Enforced on the server at the point of the call. A feature that is off in the project cannot be reached from any screen, share link or collaborator.
- A dashboard of every AI event: by feature, by user, over time, with audio minutes, words processed and the model that answered.
- The speaker change history sits beside it: every naming, rename and merge, who did it and across how many files.
- Only the project owner changes the switches. Everyone on the project can read the current setting.
How it works
Three steps from raw material to result.
Set the switches for the project
Open the project's AI settings and decide, feature by feature, what the machine is allowed to do on this film. A commissioner's condition or a participant's agreement becomes a setting rather than a memo.
The setting is enforced at the call
Every AI feature checks the project's switch on the server before it runs, and refuses if it is off. Noneditorial mode passes the check with a narrower brief: retrieve, quote, cite, nothing else.
Read the record
The dashboard lists every AI event on the project with who ran it, what it read and which model answered. Show it to the person who asked. Export the paper cut with the record beside it.
Frequently asked questions
What does noneditorial mode change?
The reading features, character overviews, connected moments, interview analysis and chat, are restricted to retrieval: they find passages, quote them and cite file, timecode and speaker, and they stop short of characterising a person or drawing an editorial conclusion. The generating features, paper cut assist, beat generation and AI voice, have no noneditorial mode; they are on or off.
Is the switch just a setting in the interface?
No. Every AI call site checks the project's setting on the server before the model is called. A feature that is off cannot be reached by a collaborator, a share link or an older browser tab.
Who can change the settings?
The project owner. Editors and viewers can see the current configuration but cannot change it, so the terms agreed for a film stay where they were set.
What is in the log?
One row per AI call: the feature, the person who ran it, the project, the model, the mode it ran in, the audio minutes or words it read, and how long it took. Transcription and translation are logged in the same way, so the record covers everything that touched the tape.
Does the log include speaker naming?
The speaker change history is a separate ledger for that: every naming, rename, merge and reset on the project, who made it and across how many files. Together the two answer the question a participant asks first, which is who decided that was me.
Does it cost anything?
No. The switches and the record are in every plan, including the free one.
Related capabilities
Further reading
Background guides and comparisons.
A grounded language model is constrained to produce output traceable to specific source chunks. Here is what grounding means in practice, why it produces usable output, and where the model can still hallucinate.
A documentary character profile gives directors a view of who they have across all interview hours. Here is what to include, what to leave out, and when in the production timeline it pays to write one.