Models

A model is a platform-side identifier (gpt-4o, sonnet-4-6, etc.) that maps to a provider, a remote model name, and the credentials needed to call it. Agents reference models by their platform name; the LLM Proxy resolves them at call time.

You need at least one LLM provider before you can register a model.

Add a model

In the Console

  1. Console → Models (/organizations/<org>/models).
  2. Click New model.
  3. Fill in:
    • Provider — pick from your registered providers.
    • Name — the platform-side identifier agents will reference (e.g. gpt-4o).
    • Remote model name — the provider's model ID (e.g. gpt-4o-2024-08-06).
  4. Save.

The model now appears in the Models list and is selectable when creating or editing an agent.

With Terraform

hcl
resource "agyn_llm_model" "gpt_4o" {
  organization_id   = agyn_organization.acme.id
  provider_id       = agyn_llm_provider.openai.id

  name              = "gpt-4o"
  remote_model_name = "gpt-4o-2024-08-06"
}

For an Anthropic model:

hcl
resource "agyn_llm_model" "sonnet_4_6" {
  organization_id   = agyn_organization.acme.id
  provider_id       = agyn_llm_provider.anthropic.id

  name              = "sonnet-4-6"
  remote_model_name = "claude-sonnet-4-6"
}

Test a model

Verifying the model works end-to-end (credentials, endpoint, model name) without creating an agent:

In the Console

  1. Models list → click the model row.
  2. In the detail pane, click Test.
  3. The Console sends Hello, world to the model through the LLM Proxy and displays the response.
  4. Errors (auth, model-not-found, rate limit) are shown verbatim from the provider's response.

The Test call hits the same LLM Proxy that agents use, so a successful test means real agents can call the model.

Use a model on an agent

Agents pick a model by name. In the Agent form, the Model dropdown lists every registered model in the organization. Set it once and it applies to every LLM call the agent makes.

Model naming convention

The model name is a platform-side identifier — pick whatever makes sense for your team:

PatternExample
Mirror the provider's marketing namegpt-4o, claude-sonnet-4-6
Use environment suffixesgpt-4o-prod, gpt-4o-dev
Use purpose-based namesfast-tier, reasoning-tier — useful if you rotate which actual model fills each role

The platform does not interpret the name beyond uniqueness within the organization.

Edit and delete

Editing a model takes effect on the next LLM call. Calls already in flight complete with the old configuration.

Deleting a model removes it from the platform. Agents that referenced it by name fail their next LLM call until you either re-create the model with the same name or change the agent's model assignment.

Token counting and cost

The platform's Metering service records every LLM call's token usage (input, cached, output, reasoning) per model. See Use → Usage for the customer view, and Operate → Monitoring for the operator view.