Tale talks to AI models through providers — each provider is an OpenAI-compatible API endpoint (OpenAI, OpenRouter, Anthropic via OpenRouter, Google, self-hosted Ollama, vLLM, etc.) together with a catalogue of model definitions. A provider exposes what models exist and how they can be used (chat, vision, embedding, image generation, transcription). Admins manage providers from Settings > Providers in the running app; users see the resulting models in the chat model picker and in agent configuration.
Tale ships with an OpenRouter example provider that gives access to models from OpenAI, Anthropic, Google, Mistral, Meta, and others through a single API key — the fastest way to get a chat workspace running end to end.
Managing providers in Settings
Open Settings > Providers. Admins can:
- Add a provider with a name, display name, base URL, API key, and one or more models. Each model entry carries an ID (must match what the endpoint accepts), a display name, an optional description, and one or more tags.
- Edit a provider to update its display name, description, base URL, API key, default models per capability, and its model catalogue.
- Delete a provider to remove it entirely. Agents that still reference the provider’s models show a warning until you pick a replacement.
The description shown in the provider list helps users understand what the provider is for (e.g. “OpenAI — Whisper for speech-to-text”). Default models per capability let you pre-select which model is used for chat, vision, embedding, image generation, and transcription when a user doesn’t pick one explicitly.
Every model belongs to one or more tags. Tags control where the model shows up in the product:
| Tag | Where the model shows up |
|---|
chat | Appears in the chat model selector and can be referenced by agents’ supportedModels. |
vision | Eligible for messages that include image attachments. |
embedding | Used by the knowledge base for document retrieval. |
image-generation | Used by image-generation agents. |
image-edit | Used by image-edit agents. |
transcription | Transcribes audio and video chat uploads — see Chat attachments. |
chat, vision, and transcription models side by side.
Making models available in chat
Providers define what models exist. Agents define which of those models they can run on. Open the agent in Agents > (agent name) and add model IDs to its model list; only models present in at least one provider and listed on the agent appear in the chat model selector.
The default chat agent is pre-configured with the OpenRouter example models. Custom agents start empty — pick the models you want the agent to support. For how the selector behaves when two providers define the same model ID (and how pinning works), see the on-disk provider reference below.
Self-hosted instances: configuration as files
Self-hosted operators can manage providers through JSON config files in addition to the UI — useful for infrastructure-as-code workflows, bulk edits, or deployments where the UI is not reachable. The UI and the files stay in sync; saving from Settings > Providers writes the same JSON.
For the file schema, bundled example providers, SOPS-encrypted secrets, self-hosted inference backends (Ollama, vLLM, LocalAI, faster-whisper-server), Docker host networking, and provider pinning syntax, see Providers — configuration reference. Last modified on April 21, 2026
