Provider API Keys

Genfeed does not generate content by itself. It calls model providers with keys that you supply. Self-hosting has no Genfeed credits: you create accounts with the providers you want, add their keys to your instance, and pay those providers directly under their own terms.

The application starts and works without any key. Add a key when you want a generation path to run.

Supported keys

All of these are optional and blank in .env.example.

Provider	Environment variable	Used for
OpenAI	`OPENAI_API_KEY`	Language models
Anthropic	`ANTHROPIC_API_KEY`	Language models
Replicate	`REPLICATE_KEY`	Hosted image, video, and other media models
fal	`FAL_API_KEY`	Hosted media models, where a media path selects fal
ElevenLabs	`ELEVENLABS_API_KEY`	Voice and audio

Other providers in the model registry (such as OpenRouter or xAI for language models) are listed in Models and providers and are configured through Integrations.

How to get each key

Create an API key in each provider’s own dashboard. Provider sites change, so follow their current instructions.

OpenAI: create a key at platform.openai.com/api-keys .
Anthropic: create a key in the Anthropic Console .
Replicate: create an API token at replicate.com/account/api-tokens .
fal: create a key in the fal dashboard .
ElevenLabs: create a key in your ElevenLabs settings .

Each provider bills you separately and may require a payment method, a funded balance, or account approval before a key can run a given model.

Add keys to your instance

Deployment-wide, in `.env`

Open the .env next to your compose.yml.

Paste each key after its variable:


OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=""
REPLICATE_KEY=""
FAL_API_KEY=""
ELEVENLABS_API_KEY=""

Recreate the stack so the container reads the new values:


docker compose --env-file .env -f compose.yml up -d

Keys in .env are server-side only. Never commit .env and never put a key in a NEXT_PUBLIC_* variable.

Per organization, in the app

Provider keys can also be entered in the app under Organization settings → Integrations. When a generation path supports it, an organization’s own key takes precedence over the deployment-wide key in .env. See AI providers for every provider the app accepts.

Check that a key works

Run a small, non-sensitive generation with a model from that provider. If it fails, check the provider dashboard for account, billing, or model-access issues, and read the container logs:


docker compose --env-file .env -f compose.yml logs genfeed

Privacy and cost

Prompts and media sent to a provider are processed under that provider’s terms. Application records stay on your infrastructure.
Genfeed does not meter bring-your-own-key usage. Provider quotas and charges apply directly to your provider account.

Running models locally

The Community image does not start GPU or local-model services, so it does not run open-source models on its own. If you run your own ComfyUI instance, point supported API flows at it with DARKROOM_COMFYUI_URL=http://your-comfyui-host:8188.