Bringing Your Own Model

FYAI includes platform models you can use straight away. Bring your own model when you need a model the platform doesn't offer, want usage billed to your own provider contract, or must keep traffic in a specific cloud account or region.

note

Bring-your-own-key models need a plan that allows them. On the standard plans that is Pro and Business. You also need the Editor role or higher in the project.

Step 1: Store the credential in an integration

The model form has no API key field. The key lives in an integration that the model links to.

1. Go to Resources → Integrations and click +.

2. Give it an Integration Name (for example OpenAI (Acme)) and a Tree Path (for example /llm-keys/openai).

3. Under Secrets, add the secrets your provider needs, using these exact names:

   Provider / auth mode	Secret names
   OpenAI, Anthropic, Google AI Studio, DeepSeek, GitHub Copilot, OpenAI-compatible	api_key
   Azure, API Key	api_key
   Azure, Client Secret (Entra ID)	azure_tenant_id, azure_client_id, azure_client_secret
   Amazon Bedrock, API Key (bearer token)	aws_bearer_token
   Amazon Bedrock, IAM Access Keys	aws_access_key_id, aws_secret_access_key, optionally aws_session_token
   Amazon Bedrock, Assume Role (cross-account)	aws_role_arn, aws_external_id
   Google Vertex AI	gcp_service_account_json (the service account key file contents)
   Ollama	None needed unless your server requires a key

   You can also store base_url or organization as secrets instead of on the model.

4. Click Create.

Use a separate integration per environment or per key so you can rotate them independently.

Step 2: Create the model

1. Go to Resources → Models and click +.
2. Enter a Model Name, for example GPT-5 Pro (Acme key). Optionally add a Description, Path and Tags.
3. Choose the Platform and API Type, then fill in the fields for your provider:

OpenAI

Field	Value
Platform	Standard (OpenAI-compatible)
API Type	OpenAI (Responses) for reasoning models, or OpenAI (Completions)
model	For example gpt-5-pro
Reasoning Effort	Responses API only: Default (unset), Low, Medium or High. Some reasoning models require High.

Leave base_url blank.

Anthropic

Field	Value
Platform	Standard (OpenAI-compatible)
API Type	OpenAI (Completions)
model	A Claude model ID, for example claude-sonnet-5-5. IDs beginning claude- are routed to Anthropic's API.

Any OpenAI-compatible endpoint

Field	Value
Platform	Standard (OpenAI-compatible)
API Type	OpenAI (Completions)
model	The provider's model ID
base_url	The API root, for example https://api.example.com/v1. Don't include /chat/completions; it's stripped if you do.

Google Gemini (AI Studio key)

Field	Value
Platform	Standard (OpenAI-compatible)
API Type	Google
model	For example gemini-3-pro
Google Search Grounding	On to let the model ground answers with Google Search.

DeepSeek

Platform Standard (OpenAI-compatible), API Type DeepSeek, and the DeepSeek model ID.

GitHub Copilot

Platform Standard (OpenAI-compatible), API Type GitHub Copilot, and the model ID. The linked integration's api_key holds the GitHub token.

Ollama

Platform Standard (OpenAI-compatible), API Type Ollama, the model name (for example llama3) and base_url set to your Ollama server. The server must be reachable from FYAI's cloud, so localhost on your own machine won't work.

Azure OpenAI or Azure AI Foundry

Field	Value
Platform	Azure (OpenAI / AI Services)
API Type	Chat Completions, Responses API (stateful), or DeepSeek (AI Foundry) for a DeepSeek deployment
Deployment Name	The deployment name in your Azure resource
Endpoint URL	Your resource endpoint, for example https://my-resource.openai.azure.com
api_version	The Azure API version, if your endpoint needs one
Auth Mode	API Key or Client Secret (Entra ID)

Amazon Bedrock

Field	Value
Platform	Amazon Bedrock
API Type	Required by the form but not used for Bedrock; choose OpenAI (Completions)
Model / inference-profile id	For example au.anthropic.claude-sonnet-5-5
AWS Region (e.g. us-east-1)	Required
Auth Mode	API Key (bearer token), IAM Access Keys or Assume Role (cross-account)

Google Vertex AI

Field	Value
Platform	Google Vertex AI
API Type	Required by the form but not used for Vertex AI; choose Google
model	For example gemini-3-pro
GCP Project ID (e.g. fyai-prod-123456)	Required
GCP Location (e.g. us-central1)	Required
Google Search Grounding	Optional

4. Under API Key Integration, click Link Integration, choose the integration from Step 1 and click Link.
5. Optionally set Context Length (total tokens) and Max Output (completion tokens) so the platform trims history correctly. Other generation settings are listed in Model Settings.
6. Click Create.

Step 3: Check it

1. Open the model's Definition tab.
2. Check the Credential card says the key is referenced from the vault, and that Findings is empty.
3. Check the Context budget leaves reasonable Prompt headroom.
4. Click Run preflight. Each check shows pass or fail with a reason.

Preflight confirms a client can be built. To confirm the provider answers, chat with an agent that uses the model.

Step 4: Use it

1. Open an agent under Build → Agents.
2. Set LLM Configuration to your new model.
3. Save, then send the agent a message from Chat or the collaboration pane.

Different agents in a team can use different models: a fast, cheap model for routing and a stronger one for analysis. To spread one agent across several models or deployments, use a model group.

Troubleshooting

Problem	What to check
Authentication errors	The linked integration has the right secret names (Step 1) and the value has no stray spaces.
404 from the provider	The model or Deployment Name is exact (they're case-sensitive), and base_url is the API root.
"Model is required when creating a non-folder config."	Fill in the model (or Deployment Name / Model / inference-profile id) field.
Model missing from the agent picker	The model saved in the current project. Switch projects if needed.
default_headers rejected	It contains a credential header. Move the credential to the integration.

Related

• Models
• Model Settings
• Model Groups and Routing
• Managing Integrations