Models¶

Argo Proxy dynamically fetches the model list from the upstream ARGO API at startup. Rather than maintaining a static list of models (which quickly becomes outdated), this page explains the naming scheme used by Argo Proxy and how to query available models at runtime.

Querying Available Models¶

CLI¶

The easiest way to see available models is the built-in CLI command:

argo-proxy models          # table format
argo-proxy models --json   # machine-readable JSON

See CLI Reference — models for details.

API¶

Send a GET request to the /v1/models endpoint:

curl http://localhost:44497/v1/models

The response follows the OpenAI-compatible format:

{
    "object": "list",
    "data": [
        {
            "id": "argo:gpt-5",
            "internal_name": "gpt5",
            "object": "model",
            "created": 1700000000,
            "owned_by": "openai"
        },
        ...
    ]
}

Each entry contains:

• id — the Argo Proxy alias you use in API requests (e.g. argo:gpt-5)
• internal_name — the upstream ARGO internal model identifier (e.g. gpt5)
• owned_by — the model provider family (openai, anthropic, google, or unknown)

Model List Refresh¶

Automatic Refresh¶

Argo Proxy periodically refreshes the model list in the background so long-running instances stay current. By default this happens every 24 hours.

You can change the interval (in hours) or disable it entirely in your config.yaml:

# Refresh every 12 hours
model_refresh_interval_hours: 12

# Disable automatic refresh
model_refresh_interval_hours: 0

Refresh events are logged at INFO level, so you can confirm they're working by checking the server log.

Manual Refresh¶

You can also trigger a refresh on-demand without restarting:

curl -X POST http://localhost:44497/refresh

See Endpoints — /refresh for details on the response format.

Model Naming Scheme¶

All Argo Proxy model names use the argo: prefix followed by a human-readable, OpenAI-style name. The naming rules vary by model family. The examples below illustrate the pattern — run argo-proxy models or query /v1/models for the actual list.

OpenAI Models¶

Standard GPT models use the format argo:gpt-{version}:

OpenAI reasoning models (o-series) have two equivalent aliases:

Both forms resolve to the same upstream model. Use whichever you prefer.

Anthropic Claude Models¶

Claude models have two equivalent aliases with different ordering:

Both forms resolve to the same upstream model. Use whichever you prefer.

Google Gemini Models¶

Gemini models use the format argo:gemini-{version}-{variant}:

Embedding Models¶

Embedding models follow OpenAI's naming convention:

Flexible Model Name Resolution¶

Argo Proxy is lenient when resolving model names. The following variations are all accepted:

• Prefix: argo:gpt-5 or just gpt-5 (the argo: prefix is optional)
• Separator: argo:gpt-5 or argo/gpt-5 (slash works as well)
• Case: argo:GPT-5 or argo:gpt-5 (case-insensitive)

Default Fallback Model¶

If a model name cannot be resolved to any known model, Argo Proxy falls back to a default model and logs a warning. This typically means the requested model name was mistyped or refers to a retired model.

Why No Static Model List?¶

The upstream ARGO API evolves over time — models are added, retired, or renamed. Argo Proxy fetches the model list dynamically at startup, refreshes it periodically, and generates aliases automatically based on the naming rules above. This means:

1. New models appear automatically — the periodic refresh picks them up without manual intervention.
2. On-demand refresh is available via /refresh or a restart if you need it sooner.
3. Documentation stays accurate without manual updates.
4. You always have the ground truth via /v1/models or argo-proxy models.