Get started
Quickstart
From signing in to your first measured model call, then asking your own assistant about it. About ten minutes.
You need a Google account and one agent or app that calls a model provider (OpenAI, Anthropic, DeepSeek, Kimi and others) with an API key. If you only use AI chat apps in the browser, follow the browser extension instead.
#1. Sign in
Open ai.odux.io and choose Sign in, then Continue with Google. Your first sign-in
creates your workspace and opens it at https://ai.odux.io/<your-workspace>/.
Note. IA4AI is in beta. If sign-in says your account cannot create a workspace, write to samstern@odux.io and we will invite you.
#2. Create your workspace key
In your workspace, open Setup and click Create your key under Your gateway key. Copy it now: it is shown
once, and IA4AI keeps only a hash of it. The key starts with ia_.
This one key is how IA4AI knows a call is yours. It belongs to your workspace only: it can send usage and read your own workspace's reports, never another workspace's. Treat it like a password. If it leaks, click Replace your key and the old one stops working at once. Your provider keys are carried over to the new key; if one could not be, Setup marks it as locked and asks you to add it again.
Then set an approval code in the card below it. Your key goes into your tools; the approval code never does. You type it only when you export captured content, replace your key or change a provider key, so a copy of the key cannot do those alone. See the approval code.
Next to the key, Open the setup guide opens your Connect page: the same steps as below, with your key already filled in and a Check my connection button after each one.
#3. Send one model call through the gateway
The gateway is the most exact way in: it forwards each call to your provider with your own provider key, and records the model, tokens, cost and tools. It never stores the prompt or the answer.
- On your Connect page, under Your agents and apps, pick your provider, paste its API key and click Test and save. IA4AI checks the key with the provider before it saves it, and stores it encrypted.
- Point your agent at the gateway, with your IA4AI key in place of the provider key:
OPENAI_BASE_URL=https://ai.odux.io/gateway/v1
OPENAI_API_KEY=YOUR_IA_KEY- Run your agent once, or send this five-token test from a terminal (use a model your provider key can call):
curl -s https://ai.odux.io/gateway/v1/chat/completions \
-H 'content-type: application/json' -H 'x-api-key: YOUR_IA_KEY' \
-d '{"model":"deepseek-chat","messages":[{"role":"user","content":"say OK"}],"max_tokens":5}'Using Claude Code or Codex? Their setup is different: see Coding agents.
#4. Check it is filed under your workspace
curl -s https://ai.odux.io/gateway/v1/whoami -H "Authorization: Bearer YOUR_IA_KEY"It answers Connected as … with your workspace. Not attributed means the request carried your provider's key instead of your IA4AI key: the call still works, but it reaches no dashboard. The fix is in step 3.
#5. See your numbers
Open Overview in your workspace: the call is there a moment later, with its cost. Agents shows each agent
separately. To give each one its own name, add x-ia-agent: <name> to its requests, or put the name in the base
URL: https://ai.odux.io/gateway/agent/<name>/v1.
Tip. A cost of
$0.00with tokens means the model had no known price when the call arrived. It is shown as unpriced, never as free, and filled in once the price is known. See How costs are counted.
#6. Ask your own assistant
Add IA4AI to Claude, ChatGPT, Cursor or another assistant as an MCP connector, then ask it:
What is my AI doing? Ask IA4AI.
The steps for each tool are in Ask IA4AI from your AI tool. For Claude it is one URL pasted into Customize → Connectors.
#Next
- Key concepts: what a workspace, an agent, a run and an investigation are.
- The gateway: naming agents, streaming, model routing and errors.
- Investigations: what you get at the end of a period.

