GuideConnect AI

Start here

Connect AI

Choose one AI route and verify a real model response. Signing in and being able to run a model are separate checks.

In the appFirst launch → Continue with ChatGPT · later: You → AI connection

On this page

Choose a connection

RouteYou provideUsage boundary
ChatGPT planAn eligible Plus or Pro account and separate consent for inference.OpenAI controls plan eligibility, visible models, capabilities, and allowances.
OpenAI APIYour own API key and an accessible API model.API usage is billed separately from a ChatGPT subscription.
My serviceYour public HTTPS service URL and its private service token.You operate the server; its provider billing and request cap also apply.

MeBot only changes to a new connection after its check succeeds. A cancelled or unsuccessful attempt keeps the previous working connection. It never switches from plan usage to API billing automatically.

Connect your ChatGPT plan

  1. In first-run setup, choose Continue with ChatGPT. To manage or change the connection later, open You → AI connection.
  2. Complete the official browser sign-in and consent flow. The local/open-source integration supports eligible Plus and Pro accounts, subject to OpenAI’s current terms and account policy.
  3. Return using Open MeBot or Android’s recent apps. Keep the app running during the attempt; if Android kills it, start a fresh sign-in.
  4. MeBot verifies the account, loads the models available to it, and completes a small model request before showing a working connection.

Your ChatGPT history stays separate.

This grant does not give MeBot your conversations or saved memory. Transfer selected preferences through the reviewed context setup flow.

Manage usage opens ChatGPT usage settings. A plan allowance, workspace policy, or unsupported capability may still prevent a request. Raising MeBot’s local request limit does not change those controls.

If the account is saved but the model fails, use Check connection, inspect Connection details, and choose another model actually returned for that account. You do not need to remove the account as the first recovery step.

What the plan connection includes

The connection uses the account’s available models for supported text inference. Web search is conditional on account/model support. MeBot keeps source links and handles an unavailable capability as a visible limitation rather than inventing current facts.

The plan route does not provide image-generation, audio, video or Files APIs, hosted MCP tools, or private connectors. Read-aloud uses the phone’s speech engine. Voice-note transcription, when offered, requires a separately configured MeBot service and is not enabled by ChatGPT sign-in. Reviewed text attachments are a separate app feature.

Provider details were checked on 23 October 2026. Read OpenAI’s current integration quickstart, models and inference guide, and preview limitations for eligibility and distribution conditions.

Use your own API key

  1. Choose Use an API key instead and enter your own OpenAI API key.
  2. In Advanced settings, keep Connection → OpenAI. Change Model if needed to one your API project can access.
  3. Tap Check and connect. This is a small real inference request and may incur API usage.

The direct-mode model field defaults to gpt-6-luna. That is an editable app default, not a guarantee of project access. API and subscription model IDs are not necessarily interchangeable. Check the official API model documentation for the capabilities you need.

Replace key changes an existing credential after a successful check. Disconnect removes the local connection. Keys stay out of context and backup exports.

Connect a personal service

This route is optional. MeBot does not include a pre-deployed public service.

  1. Deploy the supplied service behind a public HTTPS hostname, normally on port 443.
  2. In the app’s connection settings, choose My service, enter its URL and separate service token, and set the same model as the server’s MEBOT_MODEL.
  3. Run Check and connect. The phone sends the service token; the provider key stays in the server environment.

Localhost, private/LAN addresses, and custom ports are not supported for this Android connection. The service is for one person per deployment. Its independent default cap is 48 attempted calls per UTC day, and it returns completed answers rather than streaming deltas. ChatGPT OAuth tokens are never sent through this service.

See the developer setup and complete service reference.

Disconnect or change accounts

Use the app’s explicit account switch or sign-out controls. Sign-out removes local tokens and attempts remote grant revocation. If remote revocation cannot be confirmed, manage the grant through ChatGPT settings too.

Restoring a full feed backup clears local AI credentials after the data replacement and leaves background delivery off. Reconnect afterward. A context-only import preserves the connection.

Never paste an API key, sign-in callback URL, authorization code, or private response into a bug report. Share the safe error stage/code instead.

Some steps depend on the installed app release. Check the download page for the version currently offered and the verification record for known device/provider limits.

MeBot native Android screen

Native Android host renders are not physical-phone acceptance or live provider results. Check each capture’s provenance and source dates. Open original image