Fields available under Resources → Models. Labels are as they appear in the app. Several labels are the underlying API field names. Leave a field blank to use the provider's default. For concepts, see Models.
Identity
| Field | Required | Description |
|---|
| Model Name | Yes | Display name, shown in the agent LLM Configuration picker. |
| Description | No | Notes for your team. |
| Path | No | Folder in the Model Library. You can also drag the model onto a folder. |
| Tags (comma-separated) | No | Labels. |
Provider
| Field | Required | Description |
|---|
| Platform | No | Standard (OpenAI-compatible) (default), Azure (OpenAI / AI Services), Amazon Bedrock or Google Vertex AI. |
| API Type | Yes | Standard platform: OpenAI (Completions), OpenAI (Responses), GitHub Copilot, Google, DeepSeek, Ollama. Azure platform: Chat Completions, Responses API (stateful), DeepSeek (AI Foundry). Bedrock and Vertex AI are selected by Platform. |
| model | Yes, on create | The provider's model ID. Labelled Deployment Name on Azure and Model / inference-profile id on Bedrock. On the Standard platform, IDs beginning claude- are sent to Anthropic's API. |
Connection
| Field | Applies to | Description |
|---|
| base_url | Standard | API root for an OpenAI-compatible endpoint, or your Ollama server. Leave blank for OpenAI. Surrounding spaces and a trailing /chat/completions or /responses are removed. Labelled Endpoint URL on Azure. |
| Endpoint URL | Azure | Your Azure resource endpoint. |
| organization | All except Bedrock | OpenAI organisation ID. |
| api_version | Azure | Azure API version. |
| Auth Mode | Azure | API Key or Client Secret (Entra ID). Defaults to API Key. |
| azure_ad_token | Azure | A static Azure AD token (masked after saving). |
| azure_ad_token_provider (JSON) | Azure | Token-provider settings as JSON. |
| AWS Region (e.g. us-east-1) | Bedrock | Required. |
| Auth Mode | Bedrock | API Key (bearer token), IAM Access Keys or Assume Role (cross-account). Defaults to API Key. |
| GCP Project ID (e.g. fyai-prod-123456) | Vertex AI | Required. |
| GCP Location (e.g. us-central1) | Vertex AI | Required. |
| Google Search Grounding | API Type Google, Vertex AI | Off or On. Lets Gemini ground answers with Google Search. |
| Reasoning Effort | API Type Responses | Default (unset), Low, Medium or High. Only sent on the Responses API. |
| timeout | All | Request timeout in seconds. |
| max_retries | All | Retries on transient failures. |
| default_headers (JSON) | All | Extra HTTP headers. Credential headers (Authorization, api-key, x-api-key) are rejected. |
| add_name_prefixes | All | true or false: prefix messages with the speaking agent's name in multi-agent conversations. |
API Key Integration
| Control | Description |
|---|
| Link Integration / Change Integration | Opens Link Integration to Model. The model's credentials are read from the chosen integration's secrets at runtime. The linked integration's name is shown as a chip; remove it with the chip's delete icon. |
Expected secret names on the integration:
| Platform / auth mode | Secrets |
|---|
| Standard, Azure API Key | api_key |
| Azure Client Secret (Entra ID) | azure_tenant_id, azure_client_id, azure_client_secret |
| Bedrock API Key (bearer token) | aws_bearer_token |
| Bedrock IAM Access Keys | aws_access_key_id, aws_secret_access_key, optional aws_session_token |
| Bedrock Assume Role (cross-account) | aws_role_arn, aws_external_id |
| Vertex AI | gcp_service_account_json |
base_url, organization, azure_endpoint, azure_deployment and api_version may also be stored as secrets.
Context and output
| Field | Description |
|---|
| Context Length (total tokens) | The model's total context window. Drives conversation-history trimming. |
| Max Output (completion tokens) | Tokens reserved for the completion when trimming, and sent to the provider as the maximum completion length. |
| max_tokens | Maximum tokens in the response (used when Max Output isn't set). |
Generation
| Field | Description |
|---|
| temperature | Sampling randomness. |
| top_p | Nucleus sampling threshold. |
| frequency_penalty | Penalises frequently repeated tokens. |
| presence_penalty | Encourages new topics. |
| stop (string or JSON array) | Sequence or sequences that end generation. |
| seed | Seed for reproducible output, where supported. |
| n | Completions per request. |
| response_format | Output format constraint, for example json_object. |
| logit_bias (JSON) | Token ID to bias mapping. |
| user | End-user identifier passed to the provider. |
Not every provider supports every parameter.
Advanced
| Field | Description |
|---|
| model_capabilities (JSON) | Overrides for what the model supports. |
| model_info (JSON) | Additional model metadata. Google Search Grounding writes its setting here. |
Model group routing
On a group's Members tab. See Model Groups and Routing.
| Field | Default | Description |
|---|
| Group Name | | Set when creating the group (New model group). |
| Member Weight | 1 | Relative weight for the weighted strategy. Must be greater than 0. |
| Member Priority | 100 | Lower is preferred under the priority strategy. |
| Member Enabled | On | Disabled members are skipped. |
| Strategy | Weighted | Weighted (random, by weight), Priority (lowest number first), Round-robin. |
| Max failover hops / turn | 3 | 1 to 20. |
| Default cooldown (s) | 60 | 0 to 3600. |
| Fail over on | All three | Rate limit (429), Server error (5xx), Timeout. |
| Sticky (stay on last-good member) | On | |
| Honor Retry-After | On | |
| Aggressive mid-stream failover | Off | |
Actions
| Action | Description |
|---|
| + | Create a model. |
| New model group | Create a group. |
| Create / Update | Save. |
| Cancel | Discard edits. |
| Delete | Delete after confirmation. |
| Copy Model | Create a copy with the same configuration under a New Model Name. |
| Run preflight | On the Definition tab: check the configuration loads, the credential resolves and your plan permits the model. |
| Publish Template | Platform administrators only. |
Platform models are read-only, shown as Managed by platform — read only.
API-only fields and endpoints
| Item | Description |
|---|
api_key on create or update | Accepted by the API. The platform stores it in the secrets vault, links an integration named llm-<model name> and clears the key from the model record. |
GET /llmconfigs/{id}/resolved | The non-secret connection fields (model, base URL, API type, Azure endpoint, deployment and version) plus api_key_set, which says whether a credential is configured. Never returns the key. |
GET /platform-models | The enabled platform model catalog with plan floors, capabilities and per-million-token prices. No secrets. |
See Platform API.