Most "Claude trading bot" projects are weekend tutorials — a notebook, a looping script, a month of paper-trading before something quietly breaks. Thesis is the production version: Claude in the decision loop, our engineering underneath, on your broker account, managed.
A rule-based bot fires on "bullish engulfing detected" and asks no further questions. Thesis treats the pattern as a hypothesis and hands it to Claude with the context a real trader would want before committing — then accepts or rejects on that judgment, not the pattern alone.
The result is a pipeline where deterministic code does what deterministic code is good at (signal detection, sizing, execution), and a reasoning model does what rules can't (weighing ambiguous, conflicting evidence). Neither half is the product on its own.
Anyone can wire up an Anthropic API call. What's hard — and what we spent the engineering time on — is everything around it: correct position sizing on fractional shares, stop placement that survives post-entry slippage, a profit-ratchet that locks gains as they accrue, reconciliation against the broker after a disconnect, circuit breakers that pause on a bad day, and the boring backtests that tell us when a "smart" idea is actually worse.
A Claude agent you build yourself will have none of that on day one. You'll get the agent working in a weekend and then spend six months discovering why production trading is mostly plumbing.
The startups that wire Claude into every tick go out of business. Frontier reasoning is brilliant, expensive, and needs to be deployed with intent. Thesis is tuned around that constraint — the AI is called when calling it changes the outcome, and not when it doesn't.
That's the reason the monthly price is what it is. You get frontier-quality judgment on the decisions that matter, without paying for a per-trade inference bill that would sink the account.
A continuous feed of candlestick setups across the tickers you pick. Each setup is annotated with confidence, direction, and the regime it fired in.
When the AI takes or skips a trade, you see why in plain language — not just a number. Every skipped trade is logged so the behavior stays auditable.
Orders route to your broker on approval. Stops and targets are managed for the life of the trade — no forgotten open positions, no static brackets that never adjust to the move.
Switch modes any time. Run fully hands-off, approve each trade, or just watch the feed. The risk plumbing is identical across all three.
Sign up is under a minute. See the pattern feed, then decide whether you want the AI taking the shots.
Create free account →