Today (M3)
Mulno does not yet run its own AI over your data. Two things exist today that involve AI, and neither is Mulno processing your data with a model of ours:
- Cloud speech-to-text, only when a reviewer chooses it, only for the mic track, sent straight from the reviewer's machine to the studio's own account with OpenAI or ElevenLabs — never through Mulno's servers, never automatic. See the subprocessor list for what each provider does with that audio by default, and how a studio keeps ElevenLabs from training on it. Neither provider is Mulno's subprocessor: each works for the studio, under the studio's own account.
- A reviewer's own AI assistant (Claude Code, claude.ai, ChatGPT, or any other MCP client) can read and write session notes, transcripts and moments over MCP. Locally,
mulno mcpruns on the reviewer's own computer and reads the sessions there — Mulno's servers see none of it, and nothing about that use is seen or logged by us at all. Remotely (answered by our servers, for a Mulno Cloud project or a Private project with a lent key, D109): built, but not yet reachable from claude.ai or ChatGPT outside our own testing (not deployed to a public address yet). Once it is, a workspace member grants one client access to one workspace, chosen projects and named scopes (sessions:read,notes:write,media:read). The grant lasts at most 30 days before the client has to ask again, and stops working within about a minute of the member leaving the workspace; the client can revoke it, but Mulno has no screen to list or revoke grants yet. The workspace's audit log (D114) records the grant, the sign-in, every note the client writes, and every read it makes of a Private project's recordings under a lent key; its reads of a Mulno Cloud project's sessions aren't logged one by one, just as a member's own reads there aren't. Either way, that assistant runs on the reviewer's own account and plan with its provider — Mulno supplies the software, not the model, and we do not see what the assistant does with an answer.
Frames (images) are only ever handed to a client that explicitly asks for them through get_frames/get_moment — never sent anywhere automatically.
When hosted AI ships (M4, D93)
Mulno plans to sell built-in AI features (session summaries, a friction finder, build-vs-build reports, automatic tags, questions over a transcript) run on our own model account, on the following terms, which we commit to before it ships:
- Named providers. We will name exactly which model provider(s) we use here, and keep this page current when that changes (the same 30-day notice as a subprocessor change).
- Retention: zero where the provider offers it to us; otherwise the period and reason stated here, before we use that provider (owner decision, PLAN D143 O4). We will not commit in advance to a specific contract term with a provider we haven't chosen yet — that promise was withdrawn (
points-for-reviewer.mdpoint 12): a "zero retention" contract usually needs a provider's own approval and isn't available from every provider by default (some keep data briefly to detect abuse). Whichever the case, this page will say so plainly before that provider carries any of your data, not after. - No training. This one stays a binding commitment (owner decision, D143 O4): nothing sent to a hosted AI feature is used to train any provider's models — we will use only a provider whose terms with us say it won't, and we'll name those terms here. Achievable today — major providers' APIs don't train on API data by default.
- An admin's own switch, off until turned on. Hosted AI stays off for a workspace until an admin turns it on; a Private project without a lent key never sends footage to it (text only — events, notes, transcripts); a Private project with a lent key can send frames too, only if the workspace also says so.
- Bring your own key stays available for a studio whose own policy rules out our account — the same text-only rule for a Private project without a lent key applies there too.
Until this ships, no page on this product should claim hosted AI processes your data — this page will be updated the day it does, with the provider named.