Kucedr does not lock you to one built-in model. You connect the providers you want by adding your own API keys, and each capability keeps its own saved selection.
| Model service | Purpose |
|---|---|
| Assistant | The chat and agent brain. |
| Speech-to-Text | Transcription and dictation. |
| Text-to-Speech | Spoken assistant replies. |
| Text-to-Image | Image generation. |
| Text-to-Video | Video generation. |
| Music / Audio | Music and sound-effect generation. |
Scheduled work and health checks have their own provider and model selections, configured on their own Settings pages rather than during first-run setup.
Kucedr ships 31 provider manifests. The 23 model, search, and database providers covered here include Anthropic, Black Forest Labs, Brave, Cartesia, Cohere, Deepgram, DeepSeek, ElevenLabs, Google, Ideogram, Jina AI, Kimi, Kling AI, MiniMax, Mistral AI, OpenAI, Pika, Pinecone, Qwen and Wan, Reka AI, Tavily, xAI, and Z.ai. Other manifests describe MCP, storage, messaging, and local-model integrations. A catalog entry does not by itself guarantee runtime support; the tables below distinguish available, partial, and catalog-only capabilities.
Chat Routing
Kucedr uses the native Anthropic Messages API for Anthropic, the OpenAI Responses API for OpenAI, and the OpenAI-compatible Chat Completions path for every other chat provider, which is also how you connect an OpenAI-compatible endpoint of your own.
| Provider | Models |
|---|---|
| Anthropic | Claude Fable 5; Claude Opus 5; Claude Sonnet 5; Claude Opus 4.7; Claude Sonnet 4.6; Claude Haiku 4.5 |
| DeepSeek | DeepSeek V4 Pro; DeepSeek V4.1 Flash |
| Gemini 3.1 Pro Preview; Gemini 3.1 Flash Lite | |
| Kimi | Kimi K3; Kimi K2.7 Code; Kimi K2.7 Code Highspeed; Kimi K2.6; Kimi K2.5 |
| MiniMax | MiniMax M2.7; MiniMax M2.5 |
| Mistral | Mistral Large 2512; Mistral Medium 3.5; Devstral 2512 |
| OpenAI | GPT-5.6 Sol, Terra, Luna; GPT-5.5 and 5.5 Pro; GPT-5.4, 5.4 Pro, Mini, Nano |
| Qwen | Qwen3.7 Max; Qwen3.6 Plus; Qwen3.6 Flash |
| Reka | Reka Flash; Reka Edge 2603 |
| xAI | Grok 4.6; Grok 4.3; Grok Build 0.1 |
| Z.ai | GLM-5.1; GLM-5; GLM-5 Turbo |
Speech
Realtime and recorded transcription keep independent saved selections, and Settings filters models by whether they implement streaming or batch transcription. Batch audio is capped at 64 MiB of encoded input; realtime chunks at 256 KiB.
| Speech-to-text provider | Models | Modes |
|---|---|---|
| Deepgram | Nova 3; Flux | Nova 3 batch and stream; Flux stream |
| ElevenLabs | Scribe v2; Scribe v2 Realtime | Batch; stream |
| Mistral | Voxtral Mini 2602; Voxtral Mini Transcribe Realtime 2602 | Batch; stream |
| OpenAI | GPT-4o Transcribe; GPT-4o Mini Transcribe; GPT Realtime Whisper | Batch; batch; stream |
| Qwen | Qwen3 ASR Flash Realtime | Stream |
| xAI | xAI STT Batch; xAI STT Streaming | Batch; stream |
Text-to-speech input is required and capped at 4,096 characters. All seven providers below have runtime adapters.
| Text-to-speech provider | Models |
|---|---|
| Cartesia | Sonic 3.5; Sonic 3 |
| Deepgram | Aura 2 |
| ElevenLabs | Eleven v3; Eleven Multilingual v2; Eleven Flash v2.5 |
| Gemini 3.1 Flash TTS Preview | |
| MiniMax | Speech 2.8 HD; Speech 2.8 Turbo |
| Mistral | Voxtral Mini TTS 2603 |
| OpenAI | GPT-4o Mini TTS; TTS-1 HD |
Realtime Voice
OpenAI's GPT Realtime 2.1, GPT Realtime 2.1 Mini, and GPT Live 1, plus xAI's Grok Voice Latest, have runtime adapters and can power Home's realtime voice conversation with microphone input, assistant audio, transcripts, tools, and generated media in the current chat.
Media Adapters
| Status | Image |
|---|---|
| Available | Black Forest Labs: FLUX.2, FLUX.1 Kontext Pro, FLUX1.1 Pro Ultra |
| Available | Google: Gemini 3.1 Flash Image Preview, Gemini 3 Pro Image Preview |
| Available | Ideogram: 3.0, 2a |
| Available | Qwen: Qwen Image, Qwen Image Edit |
| Available | xAI: Grok Imagine Image, Grok Imagine Image Quality, Grok Imagine Image 2.0 |
| Status | Video |
|---|---|
| Available | Google: Veo 3.1, Veo 3.1 Fast |
| Available | Kling: 2.5 Turbo, 2.1 Master |
| Available | MiniMax: Hailuo 2.3, Hailuo 02 |
| Partial | Pika: 2.2. The adapter uses fal.run, so the credential and endpoint must be compatible with fal.run |
| Available | Qwen: Wan 2.5 T2V, Wan 2.2 T2V Plus |
| Available | xAI: Grok Imagine Video 1.5 |
| Status | Audio |
|---|---|
| Available | ElevenLabs: Eleven Music, ElevenLabs Sound Effects |
| Catalog only | Google Lyria 3 Pro Preview, Lyria 3 Clip Preview, Lyria Realtime |
| Catalog only | Kling Audio |
| Catalog only | MiniMax Music 2.6, Music Cover |
Only ElevenLabs currently has an audio-generation runtime adapter.
Embeddings, Search, And Vector Storage
| Area | Provider and service |
|---|---|
| Embeddings | Cohere Embed v4; Jina Embeddings v3; OpenAI Text Embedding 3 Large and Small |
| Web search | Brave Web Search; Tavily Web Search |
| Vector database | Pinecone |
A BGE-M3 adapter also exists for a self-hosted OpenAI-compatible embedding endpoint, but it has no bundled provider manifest and does not appear in the normal picker.
Save Pinecone credentials under Settings → Providers → Database, then select Pinecone explicitly in Settings → RAG. Configure the embedding provider separately under Providers → Models and approve both disclosures in RAG settings. Local retrieval uses SQLite; no vector database is selected automatically.
Credentials
Model, database, and search API keys are stored as entered in settings/providers.json and remain visible in Provider settings. Requests and credentials go to the selected provider as required for authentication and inference. Kling expects accessKey:secretKey in its API-key field. Brave and Tavily can also use their corresponding environment variables when no key is stored. See Privacy And Security for the storage boundaries.
Choosing Providers
Mix providers for quality, cost, and speed: one for assistant chat, another for transcription, another for images. Because every capability keeps its own selection, changing one does not disturb the others.