Reason like a trader.
React like a machine.
Your agent already reads the tape better than the desk. Kestrel lets it act on that judgment in milliseconds — see the market as text, plan the day, and let the runtime fire at the tick.
kestrel.markets runs it for you — the data, the authority, the deterministic execution. You bring the intelligence.
Illustrative session — generic instruments (IDX / IDXW), sample numbers.
THE PROBLEM
You connected a capable model to your brokerage. It reasons about the tape better than most humans — and still loses the trade.
Neither failure is intelligence. Both are interface.
Raw JSON and screenshots are unusable market pictures for a model. So a human ends up watching the chart and narrating candlesticks in prose to a machine — a pair of eyes bolted onto an intelligence that should already have them.
Every trading API assumes the decision-maker is fast enough to sit in the loop. An LLM is not, and will not be. By the time it requests quotes and composes an order, the market that justified the trade is gone.
Kestrel fixes the interface.
Four kinds of statement.
The whole grammar fits in a few hundred tokens — an agent holds it in context. No API docs, no SDK spelunking.
The chart, in text. Selects panes at a token budget; the live tape is vertical, so an agent pays only for new bars.
Events, not polling. The runtime watches all day; your context wakes only when something actually moves.
Judgment in advance, fired at the tick. You author bounded strategy; the runtime executes in milliseconds and wakes you in parallel.
Honest, counterfactual evaluation on date-blinded days — weights leak what code fences can’t stop.
One Frame. Every pane
your agent needs to decide.
A View selects panes at a token budget. Each moment it materializes into a typed Frame — the same numbers, rendered as text for your agent and a chart for you. It never invents a value; it selects.
A black-box recorder for agentic trading.
Every number your agent sees is typed, attributed, and watermarked — a renderer lays values out but can never invent one. Every claim gets judged, including the agent's: replayed under a pinned fill model, graded only on date-blinded days.
This is the proof your human sees at the moment of conversion — a signed Blotter and Grade, not a pitch. Self-hosted Kestrel computes the same numbers; only the platform mints the trusted receipt.
Two meters. No subscriptions.
No seats. No subscription before first value. Your agent becomes a user before you become a customer.
A trial capability, issued before any account exists.
A grade, a sim batch, a data slice — the exact amount on every request.
Always-on presence, metered against a funded balance. Budget is an Envelope field.
Bring the intelligence.
We'll handle the milliseconds.
One command, free to start. No signup, no key — prove the value first.