How it works

Your application. Our API. Your choice of model.

Ingram Cloud sits between your application and its model providers. It handles model access and usage tracking, with a hosted runtime for tools, memory, and retrieval.

Start building Read the API docs

Connect your application

Use an OpenAI-compatible SDK for Chat Completions, Responses, and embeddings. Create a project and token, then follow the quickstart for the required runtime configuration.

For chat requests, a published agent holds default instructions, model, and tools. Requests identify the application user; Ingram Cloud calls that user’s runtime instance a smith. The quickstart covers this setup.

Make your first request

Select models and credentials

Choose a model per request. Use Ingram-managed access or configure your own provider keys at project or user level.

OpenAI, Anthropic, and Google models share the chat interface. Embeddings use the OpenAI-compatible embeddings endpoint. The live catalog lists available models.

Models and keys

Configure tools and state

Keep a client-side tool loop in your application, or connect MCP tools for the platform to execute. Add conversation persistence, memory, and document retrieval as needed.

Plain Chat Completions requests are stateless by default. A thread header enables server-side conversation history. The API guide explains how request-level tools interact with hosted tools and memory.

Review integration behavior

Inspect and control usage

Review runs, model calls, tool results, token counts, and model costs in the console. Set budgets at the scopes your application needs.

Use the same data through the API and subscribe to events with webhooks. Your application can receive approval requests and submit decisions without using the console.

Inspect runs and traces

Connect your application.

Create a project, get a token, and follow the quickstart for your first API request.

Start building Read the docs