IA4AI documentation
IA4AI shows a company what its AI does and what it costs, from usage records, never from the prompts. These pages cover connecting your tools, reading the results, and how your data is handled.
IA4AI (Internal Affairs for AI) watches the two places a company's AI spend happens: the agents and apps that call models through an API, and the AI chat apps people use in the browser. It records usage: which model, how many tokens, what it cost, which tools ran. By default it never keeps a prompt, an answer, a file or a tool argument.
Each period ends in an investigation: the findings, what they cost, and the one thing worth changing next. Every number in it traces back to a measured record.
#Start here
- Quickstart
Sign in, create your workspace key, connect one tool and see your first numbers in about ten minutes.
- Key concepts
Workspaces, keys, agents, runs, investigations and findings: the words used everywhere else.
- Security
What is stored and what never is, where it runs, and what is still to come.
#Connect your AI
| You want to… | Use | In the live path? |
|---|---|---|
| Ask "what is my AI doing?" from Claude, ChatGPT, Cursor or another assistant | The MCP connector | No |
| Measure every model call exactly: model, tokens, cost, tools | The gateway | Yes |
| Measure Claude Code or Codex | Coding agents | Your choice |
| Send usage from an agent that already exports traces | OpenTelemetry | No |
| See which AI chat tools people use in the browser | The browser extension | No |
"In the live path" means your calls go through IA4AI on their way to the provider. Read If IA4AI is down before you choose.
#Use the results
- Investigations and conclusions: how a period becomes findings, and how you react to them.
- How costs are counted: what "measured" and "estimated" mean, and why an unknown price is never shown as free.
- Capture, golden sets and verification: how a cheaper model is proven on your own traffic before anyone switches.
#Build on it
- API reference: every client-facing endpoint, with its parameters.
- MCP tools: what your assistant can ask IA4AI.
- Gateway endpoints and headers: the OpenAI- and Anthropic-compatible routes.

