Skip to content

Guide / Checked Nov 2026

Is it safe to let an AI place trades?

Only with limits the agent can't change. An AI agent can send a wrong or duplicate alert, and you're responsible for every order it causes. Nyria keeps the agent out of sizing: the bot sets size from Budget per entry, the strategy's instrument, direction and trading window rules still apply, repeats send once, and a preview places nothing. Start on a paper bot, set the send tool to Needs approval in Claude, and rotate the token to cut it off.

Official documentation

Put limits around the agent

  1. 1

    Start on a paper bot

    Deploy the first bot on Nyria Paper. A wrong alert there costs nothing. Keep it on paper until the Logs tab shows a clean stretch.

  2. 2

    Let the bot set size

    Set Budget per entry on the bot. An alert names the trade and the side; the bot decides how much. The agent has no sizing tool.

  3. 3

    Tighten the strategy rules

    Set allowed instruments, direction and trading hours on the strategy. An alert the endpoint accepts can still be refused by these rules.

  4. 4

    Make sends need approval

    In Claude, open the connector's Tool permissions and set nyria_send_alert to Needs approval or Blocked. In Claude Code, leave it off --allowedTools. OpenAI says ChatGPT may ask for confirmation; don't rely on it.

  5. 5

    Preview before every send

    Ask the agent to preview. A preview places nothing and shows how Nyria read the text. Test new wording in the alert tester too.

  6. 6

    Know the two cut-offs

    Rotate token on the Integration tab stops the agent at once. Pause bot stops orders but closes nothing; if you need to be flat now, close the position at your broker.

What Nyria refuses or limits

  • Repeats send once. The same alert twice in a row is ignored as a duplicate.
  • Too many alerts are refused. A steady stream passes; bursts don't.
  • One message per request. Batched requests are refused.
  • Live accounts take market and limit orders. A stop in an alert is refused at a live account, never turned into a market order.
  • Eight tools in all. One sends and seven read. None creates bots, changes settings or rotates the token.

What's still on you

  • Nyria's disclaimer states that AI agents may produce incorrect, duplicate, or unintended alerts, and that you are responsible for configuring and monitoring any agent that sends alerts to your strategies, and for all orders those alerts generate.
  • Every active bot on the strategy gets the alert, paper and live, and accounts that copy the strategy too.
  • Nyria attaches no stop order to a live entry. Exits come from exit alerts.
  • The URL is the key. Anyone with it can send to that strategy. Keep it out of shared prompts and out of git.
  • Read the strategy and bot Logs tabs after every session.

Questions

What happens if the agent sends a wrong alert to a live bot?
The bot places it, sized by Budget per entry, within the strategy's rules. Send an exit alert or close at the broker, then read the log to see what the agent sent.
Does pausing the bot close my position?
No. Pausing closes nothing, and a paused bot ignores exit alerts too. Close at the broker if you need to be flat now.
Can the agent spend more than I set?
Each entry stays within the bot's size setting for each ticker in the alert, so an alert naming two tickers can spend the budget twice. The agent can also send more entries than you'd like, so keep sends on Needs approval until you trust it.
Does Nyria confirm with me before an order?
No. Nyria places what the strategy accepts. The confirmation step lives in the client. Claude has Needs approval, and ChatGPT may show a prompt, per OpenAI.
Is the token safe to paste into a chat?
No. Anyone with the URL can send to the strategy, so keep it out of chats and shared prompts. Paste it only into the client's connector setup. If it leaks, Rotate token invalidates it at once.

Next

Try a paper bot first

No card required. Paper is free. Live trading from $15 a month.