FAQ

Questions, answered.

The practical things to know before you install Ginu, connect another device or hand over work.

Setup & compatibility

Before the first conversation.

What do I need installed?

Invited beta testers install Ginu from the installer link we email them. You need one supported CLI installed and signed in — Claude Code, Codex or Gemini. git is required, because memory, projects and Work are stored in a git vault. tmux 3.0 or newer is required only for Shell and the in-app CLI sign-in; without it the rest of Ginu works and Shell says why it cannot open.

The first run downloads a few gigabytes of speech models. Earphones are recommended, with a built-in or wired microphone rather than using a wireless headset for both input and output. See the getting-started guide for the full sequence.

Which coding CLIs does Ginu support?

Claude Code, Codex and the Gemini CLI each have a live bridge today, and DeepSeek runs on the Claude binary pointed at its own endpoint, so it needs an API key rather than a CLI sign-in. You pick one in Settings › Engine, and only the ones Ginu can actually drive are offered.

Ginu uses the account you already have. It does not provide model access, resell the agent call or add a voice charge, and choosing an engine inside Ginu does not sign you in to that provider.

Does Ginu run on Windows or Linux?

Windows, yes: a Windows beta installer goes to invited testers. Linux is not available yet. The download page is the current source for platform availability.

Does Ginu work without an internet connection?

The voice half does: recognition and synthesis both run on your computer, with no speech-server round trip. The thinking half does not — your selected CLI sends the transcribed text and its context to its provider, so a conversation needs a connection. Setup also downloads the speech models once, and optional features such as phone pairing and extension updates make their own connections.

Data & privacy

What stays local. What does not.

What leaves my computer?

Text, and only text. Your selected CLI sends the transcribed turn and the context it needs to its own provider: Claude Code to Anthropic, Codex to OpenAI, Gemini to Google. The CLI may also read files its own configuration and permissions allow. Ginu’s operator has no server anywhere in that path, and no audio leaves the machine.

Setup model downloads, extension hosts, the daily update check and an optional phone connection make separate requests described in the privacy notice.

Where are transcripts and memory stored?

The default application data root is ~/.Ginu. Memory and project files live under its vault directory; transcripts use the configured transcript directory. These are local files you can inspect. Ginu restricts the directories to the owner, but it does not encrypt them at rest—FileVault or another disk-encryption setting is a separate protection.

Does Ginu save my raw voice recordings?

Not normally. Raw utterance audio is not kept unless saving it is explicitly switched on, which it is not by default. Text transcripts, continuity notes and diagnostic logs are separate local data.

Does this site use cookies or analytics?

No. The site sets no cookies, runs no analytics and serves its fonts, images and scripts from its own origin. It has no third-party page loads.

Using Ginu

Away from the desk. Across conversations.

What does the phone connection do?

It puts the session already running on your computer into your phone’s browser — the same conversation, work and tasks, with your computer still doing everything. You open a window for a set duration, scan its QR code and enter the passcode; five wrong guesses end that window deliberately, and a restart needs a fresh scan.

Any phone can take it, because there is nothing to install. Native iPhone and Android apps both exist as thin viewers for this connection, but neither is published, and neither is needed for this route.

The connection goes directly between your devices where possible. If a direct path is unavailable, the relay forwards encrypted traffic and cannot read its contents. See Connect your phone.

How does memory work, and can I correct it?

Memory is a tree of notes in files on your computer: what Ginu concluded about your work, not a transcript of it. A note it inferred without checking is labelled assumed, and each one keeps the belief it replaced. Correct one by saying what is wrong and what replaces it, or ask Ginu to forget it. It is not a promise that every detail of every conversation comes back.

The dreamer can review session logs and work context for repeated conventions and useful decisions. You choose when to run it, approve the operation before it rewrites memory, and inspect the result afterwards. See the memory guide.

The beta

During the beta.

Can I use Ginu for commercial work?

Yes. Commercial use is allowed during the beta. See Closed beta and the terms.

What about agent and voice costs?

Your CLI account stays yours and is billed by its own provider. Ginu does not resell it, mark it up, or add voice-minute charges — speech runs on your machine, so there are no minutes to bill.