This document is the user-facing model catalog for Friday. It summarizes the model families Friday currently exposes, which providers own each model list, and where runtime support is implemented versus still placeholder-backed or pending.
The source of truth is:
- Model catalogs and provider-keyed placeholder catalogs in
src/shared/provider-models.ts. DEFAULT_AGENT_MODELS_BY_PROVIDERandDEFAULT_PROVIDERSinsrc/shared/providers.ts.- Module model constants and runtime status in
src/shared/service.ts. - Provider adapter routing in
src/main/provider/factory.ts. - Module services, task handlers, and tool wrappers under
src/main.
Support Levels
Friday uses several kinds of model support:
| Support level | Meaning |
|---|---|
| Explicit model catalog | Friday stores concrete model ids for a provider and can validate saved selections against that list. |
| Provider-keyed placeholder catalog | Friday exposes provider/model settings for a capability, but the model id is a placeholder until provider-specific catalogs and adapters exist. |
| Endpoint-backed module | Friday can run through a configured endpoint instead of a provider/model adapter. |
| No default catalog | Friday has the store shape and target module contract, but no selectable default provider/model entries yet. |
Current module status:
| Module | Store key | Support level | Runtime status |
|---|---|---|---|
| Large language model | llmAgent |
Explicit model catalog | Implemented |
| Speech to text | speechToText |
OpenAI explicit model plus provider placeholders | Implemented for OpenAI realtime |
| Text to speech | textToSpeech |
ElevenLabs explicit model plus provider placeholders | Pending runtime |
| Text to image | imageCreator |
Provider-keyed placeholder catalog | Service, task, and tool path exist; provider adapters pending |
| Text to video | textToVideo |
Provider-keyed placeholder catalog | Pending runtime |
| Text to audio | textToSound |
Provider-keyed placeholder catalog | Pending runtime |
| OCR | ocr |
Endpoint-backed now; placeholder model constant for future provider mode | ocr.run endpoint path implemented; provider runtime pending |
| Embedding | embedding |
No default catalog | Pending runtime |
LLM Agent Models
The LLM agent uses DEFAULT_AGENT_MODELS_BY_PROVIDER. Settings return only the
static catalog for known catalog-backed providers. Known providers without a
main-agent catalog return an empty model list for the assistant. Unknown
provider ids are rejected.
| Provider id | Provider | Runtime adapter | Supported model ids |
|---|---|---|---|
openai |
OpenAI | Native OpenAI Responses adapter | gpt-5.5, gpt-5.5-pro, gpt-5.4, gpt-5.4-pro, gpt-5.4-mini |
anthropic |
Anthropic | Native Anthropic Messages adapter | claude-opus-4-7, claude-opus-4-6, claude-sonnet-4-6, claude-sonnet-4-5, claude-haiku-4-5 |
google |
Google DeepMind / Google | OpenAI-compatible chat adapter | gemini-3.1-pro-preview, gemini-3-flash-preview, gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite |
meta |
Meta | OpenAI-compatible chat adapter | llama-4-maverick, llama-4-scout, llama-3.3-70b |
xai |
xAI | OpenAI-compatible chat adapter | grok-4.3, grok-4.3-fast, grok-code-fast |
mistral |
Mistral AI | Native Mistral adapter | mistral-large-2512, mistral-large-latest, mistral-medium-2604, mistral-medium-latest, mistral-medium-2508, mistral-small-2603, mistral-small-latest, ministral-14b-2512, ministral-14b-latest, ministral-8b-2512, ministral-8b-latest, ministral-3b-2512, ministral-3b-latest, magistral-medium-2509, magistral-medium-latest |
deepseek |
DeepSeek | Native DeepSeek adapter | deepseek-v4-pro, deepseek-v4-flash |
qwen |
Alibaba / Qwen / Wan | Native Qwen adapter | qwen3-max, qwen3.5-plus, qwen3.5-flash, qwen3-coder-plus, qwq-plus |
kimi |
Moonshot AI / Kimi | OpenAI-compatible chat adapter | kimi-k2.6, kimi-k2.5, kimi-k2, kimi-latest |
zai |
Z.ai / Zhipu AI | OpenAI-compatible chat adapter | glm-5.1, glm-5, glm-4.6, glm-4.5v, glm-z1 |
minimax |
MiniMax | OpenAI-compatible chat adapter | minimax-m2.7 |
luma |
Luma AI | OpenAI-compatible chat adapter | uni-1 |
reka |
Reka AI | OpenAI-compatible chat adapter | reka-core, reka-flash, reka-edge |
perplexity |
Perplexity | OpenAI-compatible chat adapter | sonar-reasoning-pro, sonar-pro, sonar-deep-research, r1-1776 |
LLM reasoning effort:
| Provider | Saved effort behavior |
|---|---|
openai |
Saved and passed to the Responses API as reasoning.effort. gpt-5.4-mini excludes minimal; other configured OpenAI models allow none, minimal, low, medium, high, and xhigh. |
| Other providers | Saved model data is reduced to { id, name }; effort is not saved or passed by the default main-agent service path. |
Speech-To-Text Models
Speech to text currently has one explicit provider/model entry:
| Provider id | Provider | Model id | Display name | Runtime notes |
|---|---|---|---|---|
openai |
OpenAI | gpt-realtime-whisper |
GPT Realtime Whisper | Uses the OpenAI realtime adapter for live dictation and transcription. The configured model id represents the transcription model; the realtime socket may use a separate OpenAI realtime connection model internally. |
Other speech-to-text capable providers return the placeholder model id
speech-to-text-provider-coming-soon through
SPEECH_TO_TEXT_MODELS_BY_PROVIDER until provider-specific model catalogs and
runtime adapters are added: google, xai, mistral, qwen, elevenlabs,
and deepgram.
Text-To-Speech Models
Text to speech is modeled as a future module. The current code exposes one concrete ElevenLabs selection and placeholder selections for other text-to-speech capable providers:
| Provider id | Provider | Catalog model id | Runtime notes |
|---|---|---|---|
elevenlabs |
ElevenLabs | rachel-multilingual |
Concrete catalog entry, but runtime status is pending-runtime. |
openai |
OpenAI | text-to-speech-provider-coming-soon |
Placeholder catalog entry. |
google |
Google DeepMind / Google | text-to-speech-provider-coming-soon |
Placeholder catalog entry. |
mistral |
Mistral AI | text-to-speech-provider-coming-soon |
Placeholder catalog entry. |
minimax |
MiniMax | text-to-speech-provider-coming-soon |
Placeholder catalog entry. |
deepgram |
Deepgram | text-to-speech-provider-coming-soon |
Placeholder catalog entry. |
cartesia |
Cartesia | text-to-speech-provider-coming-soon |
Placeholder catalog entry. |
Exact provider/model compatibility should be validated by the TTS adapter when runtime support is added.
Text-To-Image Providers
Image creation has a provider-keyed placeholder catalog today.
TEXT_TO_IMAGE_MODELS_BY_PROVIDER maps image-capable providers to the shared
image-provider-coming-soon model id until provider-specific image catalogs
exist.
TextToImageService, the text_to_image local tool wrapper, and the
image.create background task handler are implemented. The default adapter
registry is empty, so real provider execution still requires an image adapter
for the selected provider/model pair.
Providers with image capability in DEFAULT_PROVIDERS:
| Provider id | Provider | Catalog model id | Model selection status |
|---|---|---|---|
openai |
OpenAI | image-provider-coming-soon |
Placeholder model id, pending image adapter |
google |
Google DeepMind / Google | image-provider-coming-soon |
Placeholder model id, pending image adapter |
xai |
xAI | image-provider-coming-soon |
Placeholder model id, pending image adapter |
qwen |
Alibaba / Qwen / Wan | image-provider-coming-soon |
Placeholder model id, pending image adapter |
black-forest-labs |
Black Forest Labs | image-provider-coming-soon |
Placeholder model id, pending image adapter |
midjourney |
Midjourney | image-provider-coming-soon |
Placeholder model id, pending image adapter |
kling |
Kuaishou / Kling AI | image-provider-coming-soon |
Placeholder model id, pending image adapter |
luma |
Luma AI | image-provider-coming-soon |
Placeholder model id, pending image adapter |
stability-ai |
Stability AI | image-provider-coming-soon |
Placeholder model id, pending image adapter |
ideogram |
Ideogram | image-provider-coming-soon |
Placeholder model id, pending image adapter |
Expected runtime boundary:
- The image module reads
imageCreator. - It resolves
providerIdandmodelIdfrom saved module settings. - It loads credentials and base URL from
StoreService.getProviderById. - It creates a provider-specific image adapter.
- It returns normalized image result records to UI, tasks, cron, or tools.
The LLM tool wrapper, if exposed, should pass only image instructions and safe asset references. It must not accept API keys, base URLs, or raw provider records.
Text-To-Video Providers
Text-to-video also has a provider-keyed placeholder catalog today.
TEXT_TO_VIDEO_MODELS_BY_PROVIDER maps video-capable providers to the shared
video-provider-coming-soon model id until provider-specific video catalogs
and adapters exist.
Providers with video capability in DEFAULT_PROVIDERS:
| Provider id | Provider | Catalog model id | Model selection status |
|---|---|---|---|
openai |
OpenAI | video-provider-coming-soon |
Placeholder model id, pending video adapter |
google |
Google DeepMind / Google | video-provider-coming-soon |
Placeholder model id, pending video adapter |
meta |
Meta | video-provider-coming-soon |
Placeholder model id, pending video adapter |
xai |
xAI | video-provider-coming-soon |
Placeholder model id, pending video adapter |
qwen |
Alibaba / Qwen / Wan | video-provider-coming-soon |
Placeholder model id, pending video adapter |
minimax |
MiniMax | video-provider-coming-soon |
Placeholder model id, pending video adapter |
midjourney |
Midjourney | video-provider-coming-soon |
Placeholder model id, pending video adapter |
kling |
Kuaishou / Kling AI | video-provider-coming-soon |
Placeholder model id, pending video adapter |
runway |
Runway | video-provider-coming-soon |
Placeholder model id, pending video adapter |
luma |
Luma AI | video-provider-coming-soon |
Placeholder model id, pending video adapter |
stability-ai |
Stability AI | video-provider-coming-soon |
Placeholder model id, pending video adapter |
pika |
Pika | video-provider-coming-soon |
Placeholder model id, pending video adapter |
Expected runtime boundary:
- The video module reads
textToVideo. - It resolves
providerIdandmodelIdfrom saved module settings. - It loads credentials, base URL, and webhook configuration from
StoreService.getProviderById. - It creates a provider-specific video adapter.
- It starts an async provider job, then polls or receives completion.
- It returns normalized video result records to UI, tasks, cron, or tools.
The LLM tool wrapper, if exposed, should pass only prompt, duration, aspect ratio, and safe reference asset data. It must not accept API keys, base URLs, webhook secrets, or raw provider records.
Text-To-Audio Providers
Text-to-audio and music generation use the root textToSound settings key and
the MUSIC_CREATOR_MODELS_BY_PROVIDER catalog. The catalog is provider-keyed
but placeholder-backed with music-provider-coming-soon.
Providers with sound or music capability in the default provider catalog:
| Provider id | Provider | Catalog model id | Model selection status |
|---|---|---|---|
google |
Google DeepMind / Google | music-provider-coming-soon |
Placeholder model id, pending sound adapter |
minimax |
MiniMax | music-provider-coming-soon |
Placeholder model id, pending sound adapter |
elevenlabs |
ElevenLabs | music-provider-coming-soon |
Placeholder model id, pending sound adapter |
kling |
Kuaishou / Kling AI | music-provider-coming-soon |
Placeholder model id, pending sound adapter |
stability-ai |
Stability AI | music-provider-coming-soon |
Placeholder model id, pending sound adapter |
suno |
Suno | music-provider-coming-soon |
Placeholder model id, pending sound adapter |
OCR And Embedding
OCR currently supports the endpoint-backed ocr.run task path. Provider-backed
OCR settings can use ocr with mode: 'model', but the only model constant is
document-reader-provider-coming-soon until an OCR provider catalog and
adapters exist.
Embedding has the root embedding settings shape for future semantic indexing,
but EMBEDDING_MODELS_BY_PROVIDER is empty. Do not show embedding provider
model choices until provider catalogs, vector index behavior, and runtime
adapters are implemented.
Selection And Validation Rules
- Store only
providerId,modelId, and safe non-secret options in module settings. - Keep API keys and provider base URLs on provider records in
modelProviders. - Validate saved LLM selections against
DEFAULT_AGENT_MODELS_BY_PROVIDERwhen a provider has a static catalog. - Validate speech-to-text selections against
SPEECH_TO_TEXT_MODELS_BY_PROVIDER. - For image, video, and sound, distinguish placeholder catalog selection from runtime readiness; validate adapter availability before sending prompts or assets to a provider.
- For OCR, endpoint-backed execution must use the configured endpoint; model mode must resolve a provider/model adapter before sending document data.
- For embedding, no default provider/model selections should be exposed until the catalog is populated.
- Do not duplicate provider records in task, cron, channel, or tool payloads.
Related Documentation
- providers.md documents provider credentials, API setup links, and runtime adapter behavior.
- large-language-model.md documents LLM agent execution and provider/model resolution.
- text-to-image.md documents text-to-image module boundaries.
- text-to-video.md documents text-to-video module boundaries.
- store.md documents the target model-backed store shape.