Set up Gemini and other AI services in blindGram Assistant

Connect blindGram Assistant to Google Gemini, OpenRouter, OpenAI, Claude, or a custom service. Get and paste your API key, choose a model, then enable the AI features you want.

Everything stays off until you switch it on, because each feature sends something off your device.

The first row is Conversations, your saved answers — see Conversations.

The service

These rows sit above the features and are shared by all of them.

• Provider — who you are using: OpenRouter, OpenAI, Claude, Gemini, or Custom. Choosing Custom opens the address row immediately.
• API key — your own key, pasted in. It is stored on this phone and travels with each request; it is not kept by the Assistant or sent anywhere else.
• Provider address — the endpoint. Shown for Custom only; the others fill it in for you.
• Model — the model to use. The picker offers ready-made choices, Other model… for typing an id by hand, and Load models from provider, which fetches the provider's own catalogue and gives you a search field to narrow several hundred entries.
• Speech model — the remote recognition model used by voice summaries. This row appears only when the selected provider has speech recognition. Other providers use the voice model installed on your device.

Each feature can also have Use a different service, which gives that one feature its own provider, key, and model. It is off by default — one key to paste is what most people want — and switching it on opens the provider picker straight away.

Getting an API key

You need an account with one of these and a key from it. A key is a long secret string: treat it like a password, and paste it rather than reading it out.

OpenRouter

One account reaches many models, including free ones, which makes it the easiest place to start and the default here.

1. Sign up at OpenRouter.

2. Open your account menu, then Keys.

3. Create key, give it a name, and copy the value shown once.

4. Paste it into API key.

Free models are rate-limited and vary in quality. OpenRouter has no speech recognition, so voice summaries first use the voice model installed on your device and send only its transcript to OpenRouter.

OpenAI

1. Sign up at OpenAI.

2. Open API keys in the dashboard.

3. Create new secret key and copy the value shown once.

4. Paste it into API key.

OpenAI also has speech recognition, so voice summaries send it the recording for a more accurate transcript. It is paid: add credit to the account first, or requests come back saying the quota is exhausted.

Claude (Anthropic)

1. Sign up at Anthropic Console.

2. Open API keys in the settings.

3. Create key and copy the value.

4. Paste it into API key.

Claude describes images and summarizes text well. It has no speech recognition endpoint, so voice summaries use the voice model installed on your device and send Claude only the transcript.

Gemini (Google)

1. Open Google AI Studio and sign in with a Google account.

2. Choose Get API key, then create one.

3. Copy the value and paste it into API key.

Gemini has a free tier. It has no speech recognition endpoint, so voice summaries use the voice model installed on your device and send Gemini only the transcript.

Custom

Any service that speaks the OpenAI chat format: a self-hosted gateway, a company endpoint, a proxy. Choose Custom, enter the full address in Provider address, then the key and the model id.

Describe image

• Detailed description — off, describing a photo gives you the quick answer from the model on the phone: a few object names, offline, instantly. On, the picture is also sent to your AI service, and full sentences arrive as the next message in the conversation. Switching it on shows a one-time notice naming where the picture goes.
• Description language — the language descriptions are written in. It follows the app language until you choose another.

The quick on-device answer is never replaced by an error: if the service cannot answer, the reason is added under it.

Voice summary

• Summarize with AI — adds Summarize voice to the menu of a voice message. The recording is written out and the transcript is sent to the selected provider for a summary.
• Speech model — the remote recognition model used for the first step. It appears only for providers with speech recognition.

Every provider can summarize voice. OpenAI receives the recording. A Custom gateway tries remote recognition first and falls back on-device when that endpoint is unavailable. OpenRouter, Claude, Gemini, and other providers without speech recognition use the installed on-device voice model; if its language model is missing, the app offers the download and continues the same summary afterwards.

Text summary

• Summarize messages — adds Summarize text to the menu of a long message, from about 200 characters up. Below that, the message is quicker to hear than its summary.
• Summary language — the language summaries are written in. Left alone, a summary is written in the language of the message itself.

Ask about messages

• Ask AI about any message — adds Ask AI to the message menu, for any message at all. The message opens as a conversation and you type your own question.

What is sent: the message written out, and the attachment itself — a picture as a picture, and anything with sound written out first, which costs one extra recognition request. A file not yet on your phone is downloaded before the question is answered.