Privacy

Where your data goes

The complete map: what stays on your machine, what touches maneku's servers, and what the AI providers see. Written for people who read this stuff carefully, because we would too.

The short version. Your recordings, transcripts, and everything said in your room stay on your machine. Your maneku account syncs presets and session stats, never content. On a bring-your-own-key plan, AI traffic goes straight from your computer to your providers with your keys. It does not pass through maneku's servers.

What never leaves your machine Local

Everything below is written to a folder on your own disk (inside the app's data directory) and has no upload path in the app. That's not a policy promise, it's how the software is built: there is no code that sends these anywhere.

Want to check? Every session folder contains human-readable files: transcript.txt, session.json, and small CSV files of usage numbers. Open them. What you see is everything there is.

What syncs to your maneku account Cloud

Signing in is what makes your setup follow you across devices and lets us check your plan. Here is the complete list of what our servers store for you:

DataWhat exactly
AccountYour email address and sign-in method.
PersonasName, description, voice and model choice, and personality sliders for each persona you create.
ShowsShow name, default persona, and any standing show text you save (like a pre-show brief or a custom prompt).
SettingsFour preferences: voice provider, two budget alert numbers, and your summon hotkey.
Session statsWhen a session started and ended, how many summons, interrupts, chats, and greetings happened, estimated cost, and which show and persona ran. Numbers and names only.

Not on that list, and not stored by our servers: transcripts, audio, summon content, episode notes, reference materials, API keys. Nothing derived from your transcript syncs either.

The app's other network traffic

That's the whole list. If the app is closed, nothing runs and nothing is sent.

What the AI providers see

On a bring-your-own-key plan, your machine talks directly to three providers using your own accounts. maneku is not in that path. What each one receives, and what they say they do with it:

Anthropic (the co-host's brain)

Receives the recent conversation transcript plus your persona and show context, and returns the co-host's response. Anthropic states it does not train on API data by default and automatically deletes API inputs and outputs within 30 days (longer only for abuse-flagged content). Sources: training policy, retention policy.

Groq (transcription)

Receives short chunks of room audio and returns text. Groq states it does not retain inference data by default (up to 30 days only for reliability and abuse monitoring, which you can switch off entirely with the Zero Data Retention setting in your Groq console), and its services agreement prohibits training on your data. Sources: data practices, services agreement.

ElevenLabs (the voice, optional)

Receives the co-host's response text and returns audio. This is the one to configure: by default ElevenLabs keeps your generations in your account history until you delete them and may use submitted content to improve its models unless you opt out. The opt-out lives in your ElevenLabs account under Terms and Privacy → Data use. We recommend flipping it. Sources: privacy policy, retention docs.

On a credit plan it works differently. Credit-plan summons route through maneku's server, because that's where credits are counted and the provider keys live. On that path your summon content passes through our infrastructure, in memory, on its way to the providers. It is relayed immediately and never written to disk, logs, or the database — the audit covered this path too, and the relay code records usage counts only. Bring-your-own-key traffic never touches our servers at all.

How we know this is true

This page was written from a code-level audit of the shipping build, not from intentions. We enumerated every network call in the app and every file it writes. Where we found daylight between our claims and the code, we changed the code: as of this build, session sync sends numbers and names only, with nothing derived from your transcript.

Last verified against the shipping build on July 22, 2026. Found something that doesn't match? Tell us: hello@maneku.ai. We'd genuinely like to know.