Image

Pictures for buddi. An agent makes one image from a prompt, optionally from pictures you hand it, and keeps it in your Files library, where the canvas shows it.

For an agent your other agents can ask for pictures, add the Illustrator from buddi's agent catalogue. If you accepted it (@art) from an earlier version, it keeps working as it is, and buddi offers it the catalogue's updates while you have not edited it.

What it needs

An image service you already have, added on Settings → Model accounts: an OpenAI account, a ChatGPT subscription, Gemini, or any service with an OpenAI-compatible images endpoint (a local or hosted FLUX server, for instance). Settings → Image lists the accounts it can use. buddi never reads a key from your environment.

What it costs

The plugin is free. Each picture is billed by the service you chose, at its price (per image on an API account; against your plan on a subscription). At most 30 images a day by default; you change the limit on Settings → Image.

What leaves your computer

The prompt, and any reference pictures, go to the service of the account you chose. The picture comes back and stays on your computer. Nothing else is sent.

What it asks before doing

The first image in a conversation is an approval card naming the account and the model; later ones in that conversation run. A colleague the conversation delegated to counts as the same conversation.

Install

buddi plugins install @withbuddi/plugin-image

or find it in buddi's plugin market on Settings → Plugins. buddi stages it and shows what it claims (what it reaches in buddi, the hosts it talks to); nothing runs until you approve it. Then choose an account on Settings → Image, and add the Illustrator from the catalogue or give image.generate to an agent of your own.

Remove

buddi plugins uninstall image --yes

Your pictures stay in the Files library and the plugin's settings stay in the database. --purge --confirm image deletes its settings and history too; that cannot be undone.

For developers

image.generate { prompt, references?, name?, aspect? } returns { id, name, bytes, width?, height?, backend, account, model, prompt, forAgent }: the canvas folds prompt under the picture and never draws forAgent, the model's own instruction (core's AGENT_ONLY_FIELD).

Each service is one ImageBackend in src/backends/, listed in BACKENDS. The tool is declared auto and narrowed by tierFor: gated until the owner has approved one image.generate in the conversation, then auto. With no usable account chosen it refuses at once and raises no card.

pnpm install && pnpm --filter @withbuddi/plugin-image build
DATABASE_URL=postgres://… pnpm --filter @withbuddi/plugin-image test

The tests run fake services locally; with DATABASE_URL they also run the tool on Postgres (account choice, refusals, the approval, the cap).

Licensed under Apache-2.0.