INTERFACES / OTTTO IN CHATGPT

Ask Ottto inside ChatGPT.

Ottto answers in ChatGPT too. Ask about your AI coding-agent spend in plain language and get the same trusted numbers back as inline cards - cost, what changed, source health, and reviewed savings recommendations. Read-only, and never your raw prompts or keys.

Start free See what you can ask →
Ottto in ChatGPT
WHAT YOU CAN ASK

Your AI spend, in the chat you already use.

No dashboard to open, no export. Ask in plain language and Ottto returns a compact card built from your own account - the same reconciled numbers you see in the web app.

What's my Ottto account snapshot? What changed in my AI usage this week? Are my data sources healthy? Show my Ottto recommendations
ChatGPT Ottto
What changed in my AI usage this week?
Ottto cost-breakdown card inside ChatGPT: weekly spend with the top models that drove it
Show my recommendations
Ottto recommendations card inside ChatGPT: a reviewed savings recommendation
Real Ottto cards, rendered inside ChatGPT.
SAFE BY CONSTRUCTION

It can read your numbers. It can't touch your account.

The app is read-only and data-minimized at the boundary - a projector builds every payload by allowlist, so the things ChatGPT should never see never leave.

Read-only tools only

Account snapshot, what-changed-this-week, source health, and reviewed recommendations. No tool can change your account.

Data-minimized projector

Every field is opted in by allowlist. Raw prompts, file paths, device ids, emails, and API-key prefixes never cross the boundary.

You connect your own account

OAuth 2.1 with PKCE - you authorize Ottto to read your data, and you can revoke it any time.

Freshness & confidence on every answer

Each card carries how current and how complete the picture is, plus the caveats - the same labeling you trust in Ottto.

Local MCP for agents

And a second MCP - for your own agent.

Different audience, different job. The ChatGPT app above is for asking about your spend. This one is for operating Ottto from inside your coding agent: point Claude Code, Cursor, or Codex at a local ottto MCP server and ask it to check the platform's health, repair the daemon, run diagnostics, or explain why no telemetry is flowing.

  • ✓
    Read-only by default - status, doctor, and diagnostics are safe to call on any turn.
  • ✓
    Mutating actions gated - setup, fix, and restart carry destructive hints so your agent confirms first.
  • ✓
    Wraps the stable ottto --json CLI - one versioned contract over local stdio, no private internals.

Same boundary as the CLI: local, JSON-first, and explicit about actions that can mutate state.

claude-code · ottto local MCP
# your agent asks Ottto to check itself "is my local platform healthy?" → ottto status --json ✓ daemon running · 4 sources · telemetry flowing "why is no telemetry showing for codex?" → ottto doctor --app codex --json ✓ found + repaired: otel config path

See your real numbers, wherever you ask.

Connect your accounts in the web app, then ask Ottto in ChatGPT or from the coding agent you already have open.

Start free See the web app →
Read-only · data-minimized · your prompts and keys stay yours.