Skip to main content

Models

Every agent runs on a model configuration: which provider and model to call, how to authenticate, and the generation settings to use. You manage them under Resources → Models, in a tree called the Model Library, and choose one per agent in the agent's LLM Configuration field.

There are three kinds of entry in the Model Library:

KindWhat it is
Platform modelA model FYAI provides and pays the provider for. Usage is billed through your FYAI plan. Read-only.
Your own model (bring your own key)A configuration you create that calls a provider with your credentials.
Model groupA routing entry that sends each turn to one of several member models and fails over between them.

Platform models​

Platform models appear in the Model Library and in the agent model picker with a Platform chip, and show Managed by platform — read only when opened. You don't need any credentials to use them.

  • Plan floors. Each platform model can have a minimum plan. The Who can use it card on a model's Definition tab shows its Minimum plan, any plans it's hidden from, and who pays.
  • Opting out. A project created with Include platform-managed resources turned off doesn't list platform models.
  • Which models. The catalog changes as providers release and retire models. Open the Model Library, or the LLM Configuration picker on an agent, to see what your plan can use today. At the time of writing it includes models such as Azure DeepSeek-V4-Flash and GLM 4.7.

The platform catalog, its pricing and per-model rate limits are managed by FYAI platform administrators under Settings → Model Catalog. That page isn't available to customers.

Bring your own model​

You can connect your own provider account. Bring-your-own-key (BYOK) models require a plan that allows them; on the standard plans that is Pro and Business.

Provider support comes from the Platform and API Type you choose:

PlatformAPI TypeUse for
Standard (OpenAI-compatible)OpenAI (Completions)OpenAI Chat Completions, and any OpenAI-compatible endpoint via base_url (for example Mistral, Groq, OpenRouter or a self-hosted server). Model IDs beginning claude- are sent to Anthropic's API.
OpenAI (Responses)OpenAI's Responses API, including reasoning models (enables Reasoning Effort).
GitHub CopilotModels available through a GitHub Copilot licence.
GoogleGemini models through a Google AI Studio API key. Enables Google Search Grounding.
DeepSeekDeepSeek's own API.
OllamaAn Ollama server reachable from the FYAI platform.
Azure (OpenAI / AI Services)Chat Completions, Responses API (stateful), DeepSeek (AI Foundry)Your Azure OpenAI or Azure AI Foundry deployments.
Amazon BedrockModels in your AWS account, by model or inference-profile ID.
Google Vertex AIGemini models in your GCP project. Enables Google Search Grounding.

Credentials aren't typed into the model form. You link an API Key Integration: an integration whose secrets hold the key (api_key, or the provider-specific names listed in Bringing Your Own Model). If you create a model through the API with an api_key value, the platform moves the key into the secrets vault and links an integration named llm-<model name> for you.

Credential headers (Authorization, api-key, x-api-key) aren't accepted in default_headers; use the linked integration instead.

Model groups​

A model group routes each turn to one of its members and fails over when a member hits a rate limit, a server error or a timeout. Assign a group to an agent exactly like a model. Groups let you spread load across deployments, prefer a cheaper model with a fallback, or ride out a provider's capacity limits. See Model Groups and Routing.

The context budget​

Prompt and completion share one context window. On a model's Definition tab, the Context budget card shows the Window (context_length), Max output, the Prompt headroom left once the completion is reserved, and the size conversation history is trimmed to. If you set max_output close to the window, very little is left for the prompt.

Model page tabs​

TabContents
OverviewReach, spend and the model's contract at a glance.
DefinitionContext budget, Credential (where the key comes from: vault, stored on the row, Azure AD, the platform's identity, or nothing), Who can use it, Findings (configuration problems), Routes to for a group, and Preflight.
ConsumersThe agents built on the model and what each has spent.
MembersGroups only: the member and routing editor.
DetailsAll editable fields. See Model Settings.
HistorySaved versions, with rollback.

Run preflight loads the configuration the way an agent would, checks the credential resolves and asks the plan gate. Passing means a client will build; it doesn't prove the provider will answer, so test with a short chat too.

Model governance inventory​

For audit and procurement, the API endpoint GET /model-governance lists every provider and model your project can route to, with hosting region, data retention, whether the provider trains on customer data and lifecycle status, plus /model-governance/usage for which models your data has actually reached. Project administrators can record assessments with PUT /model-governance/{provider}/{model}. This is API-only; see Platform API and Governance.