Models View

Configure AI providers, model catalog, aliases, and per-agent defaults

The Models view is where you tell Comis which AI providers it can use, which specific models are available, what shorthand aliases point to them, and which model is the default for each agent.

Who it’s for: anyone deciding which models their agents call — from “swap Claude Sonnet for GPT-4o” to “configure a self-hosted Ollama endpoint.”

Route

• /models — opens the Models view

What You See

The view is organized into four tabs:

Providers tab

Lists every custom provider entry from the providers.entries config block — these are OpenAI-compatible proxies, self-hosted gateways, and other endpoints not covered by the pi-ai built-in catalog. Most users will see this tab empty, because the pi-ai catalog covers all common providers natively (see the Catalog tab below).

Each card shows:

• Type — any pi-ai catalog provider name, or openai for an OpenAI-compatible custom proxy
• Name — a friendly label
• Base URL — the endpoint the SDK calls
• API key env name — which environment variable holds the key (e.g., ANTHROPIC_API_KEY)
• Status — a connection dot (green = key present and ping succeeded; red = missing or failing)
• Timeout / Max retries — per-provider transport tuning
• Enable toggle — soft-disable a provider without removing it

A Test button on each card calls models.test to run a connectivity check and (when supported) lists which models the provider currently exposes plus which agents are using it.

The daemon includes a guard that rejects redundant custom entries for built-in providers — if you try to create providers.entries.openrouter with the catalog’s default baseUrl, the daemon returns an actionable error directing you to use gateway env_set OPENROUTER_API_KEY=<key> + agents_manage update --provider openrouter instead. Custom entries are only needed for proxies (different baseUrl) or non-catalog providers.

Catalog tab

Lists every model from the pi-ai SDK catalog — sourced live from the models.list RPC, which itself wraps getModels(provider) from @earendil-works/pi-ai. Each row shows:

Field	Description
Provider	Which provider serves the model
Model ID	Vendor’s canonical id (e.g., claude-sonnet-4-5-20250929)
Display name	Friendly label for UI
Context window	Maximum input tokens
Max output tokens	Per-completion cap
Reasoning	Whether the model exposes reasoning blocks
Validated	Whether Comis has confirmed the id resolves at the provider

A search input filters the list. Use this tab to check whether a specific model id is available before assigning it.

Aliases tab

Aliases are shorthand names you can use in agent config (e.g., model: fast) instead of the full vendor id. Each row shows:

• Alias — the shorthand
• Provider + Model ID — what it resolves to

Add, edit, or delete aliases here; changes are saved via config.patch.

Defaults tab

Shows each agent’s effective model resolution. For every agent you’ll see:

• Agent ID
• Provider the agent is currently configured to use
• Model ID — either explicit, or resolved from an alias
• Override controls if you want a per-agent model that differs from the global default

Use this tab when you want to swap one agent to a faster/cheaper model without touching the others.

Common Tasks

1. Add a new provider

   Click Add Provider on the Providers tab. Pick the type (e.g., groq), set a name, paste the API key into your environment file or use the secrets manager, then click Test to verify connectivity.

2. Find which agents use Claude Sonnet

   On the Providers tab, click Test on the Anthropic card. The test result lists every agent currently using an Anthropic model.

3. Create a 'fast' alias

   On the Aliases tab, click Add Alias. Enter fast as the alias, pick groq as provider, and llama-3.3-70b-versatile as the model id. Save. Now any agent with model: fast resolves to that combination.

4. Override one agent's model

   On the Defaults tab, find the agent you want to change. Set its provider and model. Save — the change writes to agents.<id>.provider and agents.<id>.model and triggers the daemon to reload.

This view shows the user-facing slice of model configuration. The full set of provider options (rate limits, fallback chain, cost tracking) lives in the models section of config.yaml — see the Config Editor for raw YAML access, or the models reference for the schema.

Related Pages

Models Reference

Full provider catalog, default models, and pricing.

Agent Editor

Set per-agent model overrides from each agent’s editor page.

Config Editor

Edit the raw models section of your YAML.