Agent Setup

This flow is for users who have already tried Robin in the UI, created an account, and generated a Robin API key.

The key idea: store the API key in the local robin CLI config. Do not paste the raw API key into an AI agent chat. The agent can use Robin through the local CLI.

1. Install The CLI


curl -fsSL https://storage.googleapis.com/robin-downloads/cli/install.sh | sh

Verify the install:


robin --version

2. Store Your API Key


robin auth set-key

Paste the Robin API key when prompted. The key is stored locally for the CLI.

Check that the CLI can read your account:


robin me

3. Tell Your Agent What To Read

Open Codex, Claude Code, OpenClaw, or another local agent tool, then give it a prompt like this:


Read https://docs.robin-predict.com/SKILL.md and use the local robin CLI.
You may inspect markets, portfolio state, quotes, and trades.
Ask me before executing buy or sell trades.

This default prompt lets the agent inspect Robin, reason about markets, and ask for approval before it submits a trade.

You can also give the agent your preferences, constraints, or private context:


Read https://docs.robin-predict.com/SKILL.md and use the local robin CLI.
I prefer markets about technology and macro policy. I avoid sports markets.
I am comfortable risking up to 10 RPM on a single idea.
Use outside sources before recommending a trade.
Ask me before executing buy or sell trades.

If you want to authorize limited trading, give the agent explicit limits:


Read https://docs.robin-predict.com/SKILL.md and use the local robin CLI.
You may trade up to 10 RPM per trade and 50 RPM total today.
Always quote before executing. Verify with robin trades after execution.

4. Recommended Agent Loop

A typical Robin session is conversational. The user asks for interesting markets, the agent checks portfolio state and open markets, then explains a few candidates before quoting or trading.

Start with discovery:


robin me --json
robin markets --limit 10 --json

If a market looks interesting, the agent can research outside sources, explain the thesis, then quote the proposed trade:


robin quote-buy <marketId> --outcome-idx <idx> --amount-in <amount> --json

Before execution, the agent should tell the user the side, market, outcome, amount, slippage policy, and why the trade may be attractive. If the user has clearly allowed execution, then:


robin buy <marketId> --outcome-idx <idx> --amount-in <amount> --idempotency-key <key> --json
robin trades --include-pending-by-me --limit 10 --json

For sell-side workflows, use quote-sell before sell, then verify with robin trades.

5. Interaction Modes

Suggest-only: the agent can inspect markets, portfolio state, quotes, and outside sources, but does not submit trades.
Confirm-before-trade: the default mode. The agent asks before every robin buy or robin sell.
Limited autonomy: the agent may trade without asking each time, but only inside explicit limits such as max RPM per trade, daily budget, market categories, outcomes, and stop conditions.

Safety Notes

Quote commands are read-only and do not submit trades.
buy and sell submit trades, so agents should only run them when the user has clearly allowed execution.
Use --idempotency-key when retrying buy or sell commands.
Verify submitted trades with robin trades --include-pending-by-me --json.