Docs/TypeScript/Custom Models
Configurationmodels

Custom Models

The set of models indusagi can use comes from the bundled indusagi framework registry. This page covers how to inspect that catalog, narrow which models appear in the picker, select one for a run, and register a model programmatically.

Table of Contents

Listing Models

--list-models prints the full catalog as an aligned table — provider, model id, context window, max output, whether it supports reasoning, and whether it accepts images. An optional substring filters the rows:

indus --list-models
indus --list-models claude        # only rows whose provider/model id contains "claude"

The catalog is sourced from getProviders() / getModels() in indusagi/ai. The framework's in-process mock provider (one model, mock-default) is filtered out of the interactive model catalog, so it is never the auto-selected default and never appears in the /model picker. It is not filtered from --list-models, which prints the raw framework registry, so mock-default can show up there.

Selecting a Model

Pick a model for a run with --model (alias -m). It accepts a provider-qualified id or a bare id when that id is unambiguous across providers:

indus --model anthropic/claude-sonnet-4-20250514
indus -m claude-sonnet-4-20250514

With no --model, the agent chooses a default model belonging to the first provider you are authenticated with. Inside an interactive session, /model (alias /models) opens the picker.

Narrowing the Picker

The enabledModels setting restricts which models the picker offers. It is a list of glob / id patterns; an empty list (the default) means every model is offered.

{
  "enabledModels": ["claude-*", "gpt-4o", "gemini-2*"]
}

See Settings.

Supported APIs

Each model declares the streaming api its provider speaks. The framework wires up these built-in API adapters:

API Description
anthropic-messages Anthropic Messages API
openai-completions OpenAI Chat Completions API (most compatible)
openai-responses OpenAI Responses API
azure-openai-responses Azure OpenAI Responses API
openai-codex-responses OpenAI Codex Responses API
google-generative-ai Google Generative AI
google-vertex Google Vertex AI
bedrock-converse-stream Amazon Bedrock Converse API
kimi-openai-compatible Kimi (Moonshot) OpenAI-compatible API
nvidia-openai-compatible NVIDIA OpenAI-compatible API
mock-ai In-process echo provider (tests only)

Registering a Model Programmatically

To add a model that the bundled catalog does not ship, use the framework ModelRegistry. It exposes registerCustomModel(model) to add one model and loadCustomModels(models) to add several. A registered model becomes visible to getModels() for its provider, so it shows up in --list-models and the picker.

import { ModelRegistry } from "indusagi/ai";

const registry = new ModelRegistry();

registry.registerCustomModel({
  id: "my-local-llama",
  name: "Llama 3.1 8B (Local)",
  api: "openai-completions",
  provider: "ollama",
  baseUrl: "http://localhost:11434/v1",
  reasoning: false,
  input: ["text"],
  cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
  contextWindow: 128000,
  maxTokens: 32000,
});

For routing an existing provider through a proxy, or implementing a non-standard streaming API, see Custom Providers.

Model Shape

A model is the framework Model record (from indusagi/ai):

Field Type Description
id string Model identifier (e.g. "claude-sonnet-4-20250514")
name string Display name
api string Streaming API the provider speaks (see above)
provider string Owning provider id
baseUrl string API endpoint URL
reasoning boolean Whether the model supports extended thinking
input ("text" | "image")[] Supported input types
cost object { input, output, cacheRead, cacheWrite }, $ per million tokens
contextWindow number Context window size in tokens
maxTokens number Maximum output tokens
headers object Optional custom request headers
compat object Optional OpenAI-compatibility quirks (for openai-completions / openai-responses)
← Previous
Settings
Next →
Custom Providers