algoth.ink

Docs — Automation

AI trade reviews

How AlgoThink's AI reviews work, what agree/neutral/disagree means, how to use your own API key, and why the AI never originates or executes a trade.

An AI review is a second opinion on a suggestion a strategy already produced. It reads the trade alongside the indicators and recent news and gives a plain-language agree, neutral or disagree take, with its reasoning.

What it does not do

The AI never originates or executes a trade. It only comments on signals the deterministic strategies already produced. Trade ideas come from tested rules; the decision stays with those rules and, ultimately, with you.

Turning it on

AI reviews are a Pro feature. The toggle is in the Risk & controls card on the Dashboard. Once on, reviews attach to new suggestions automatically; you can also request one on demand from a suggestion card with the Review button.

Using your own API key

Pro includes reviews using the platform's key. If you'd rather control that usage — your own rate limits, your own billing, your own model choice — add your own key under Account → API keys & notifications. Three providers are supported:

  • NVIDIA API key (AI reviews)
  • OpenRouter API key (AI reviews)
  • Anthropic API key (Claude reviews)

Your key is encrypted before it reaches the database and is only decrypted to make the review call you asked for. Usage on your own key is billed by that provider, not by AlgoThink.

When the AI is unavailable

Trading keeps working with no AI at all. If a provider is down, rate limited or unconfigured, the suggestion simply shows as having no review rather than blocking anything. Reviews are an input, never a dependency.

Requiring agreement in autonomous mode

If you use autonomous mode, one of the extra conditions you can set is require AI agreement — only opening a position when the reviewer's stance is "agree". Worth knowing: this makes the automation dependent on a system that can be unavailable, so fewer trades will fire.

What to consider

  • Treat it as a sanity check, not an oracle. It's a language model reading indicators and headlines — useful for catching context a rule set can't see, not a source of truth about where price is going.
  • Disagreement is informative. A review arguing against a signal is doing its job; the Dashboard tracks whether the AI's opinions actually improved on taking every signal blindly.
  • It sees news, so it inherits news bias. Thin coverage on smaller coins means less to work with.

Related

AI reviews come with Pro — try the platform free first.

Create free account