Guides
Step-by-step walkthroughs, from your first decision to scheduled, automated ones; then cookbooks, worked recipes with real responses; then use cases: practical patterns with ready-to-run examples.
Getting started
Your first decision in the Playground
Ask three trading questions about a stock with nothing but its symbol, read the probabilities, and turn one of them into a decision. No code and no API key.
7 steps · about 5 minCall the API from your code
Create an API key, keep it out of your code, send your first request and handle the answer, in the language you use.
6 steps · about 10 minSave, share and reuse queries
Keep the requests you use often, share them with your team, and turn any of them into code.
5 steps · about 3 minFinancial decisions
Decide on a stock with just its symbol
Ask trading questions about a US stock or ETF without sending any market data: the server fills in the latest completed session.
4 steps · about 5 minWrite your own rule
Write a rule about a stock's price history, read the belief it returns, combine it with forecasts, and act on it daily.
9 steps · about 12 minRead probabilities and choose thresholds
A probability is not a signal until you decide where to act. Here is how to pick that line and check it against what really happened.
5 steps · about 8 minRule playground
From English to rule
Ask about what is true now in plain English, see the rule Ora compiled it to, and learn when to write the rule yourself.
5 steps · about 8 minCombining conditions
Join rules with and, or and not, see what each does to the belief, and avoid the precedence trap.
7 steps · about 8 minOffsets and windows
Look back with offsets and ref(), choose between windows that include today and ones that do not, and know what happens when history runs out.
6 steps · about 8 minReading beliefs
What a rule's number means, how it grows with the margin, and the fixed points: 0.50, the 0.55 / 0.45 lean at a tie, and 0.95.
5 steps · about 8 minRules plus forecasts in one request
Ask what is true now and what is likely next about one symbol in a single call, and combine the two in code.
5 steps · about 8 minTurning a rule into an alert or workflow
Put rules and forecasts in an action so Ora posts to Slack or your webhook when they line up, then run it over a watchlist every day.
5 steps · about 10 minAutomation
Send a Slack alert when a rule fires
Answer the questions, test a rule on the answers, and post to a Slack channel only when it is true.
6 steps · about 10 minSend decisions to your own system
Deliver a signed JSON payload to your order system, risk system or automation tool (Zapier, Make, n8n) whenever a rule fires.
6 steps · about 15 minRun a watchlist every day after the close
Ask the same questions about every symbol on a list, on a schedule, and get one digest message per run.
5 steps · about 10 minProduction
Cookbooks
All cookbooks
Worked recipes for ora-fin-0.1.1, market first: each one has a goal, when to use it, the request, the real response, the code, how to act on it, and variations.
Rank a watchlist by probability
One request per symbol, sent in parallel, then rank the list by a yes_no answer.
Screening and digests · Beginner · about 10 minSend a pre-open digest
A weekday workflow: the same questions for a watchlist, one Slack message per run.
Screening and digests · Beginner · about 10 minTrend filter plus entry trigger
Rule C > sma(C,200) as the filter, a forecast as the trigger, one action when both agree.
Signals: rules plus forecasts · Intermediate · about 8 minBreakout confirmation with volume
A breakout rule with a volume condition, its two halves asked separately, and a forecast.
Signals: rules plus forecasts · Intermediate · about 8 minRSI and Bollinger mean-reversion checks
Is it stretched now (rules), and does the next week lean toward a bounce (forecast)?
Signals: rules plus forecasts · Intermediate · about 8 minGap-fill playbook to your order system
fill_today / fill_5d on your own gap numbers, three branches, a signed webhook.
Signals: rules plus forecasts · Advanced · about 12 minConfidence-gated actions
Act automatically only outside a band around 0.5; send the rest to a person.
Risk and execution · Beginner · about 6 minDetect two-way risk
drop5 and rally5 both high: a wide range, with an options angle.
Risk and execution · Intermediate · about 8 minSize positions by volatility regime
ATR-percentage and realized-volatility rules pick a size multiplier.
Risk and execution · Intermediate · about 8 minRisk guardrail before an order
Block, reduce or allow an order from drawdown odds and a trend rule, via webhook or inline.
Risk and execution · Advanced · about 12 minAsk many questions in one request
Eight questions, forecasts and rules, in one call: one round trip and far fewer tokens.
Techniques · Beginner · about 5 minSelf-consistency: three wordings
Ask the same question three ways in one request and use the spread to catch bad wordings.
Techniques · Intermediate · about 8 minRoute support tickets
choice + score + yes_no on a ticket, routed only when confidence is high.
Text · Beginner · about 5 minIs a headline relevant to a symbol?
Relevance, tone and materiality of a headline for one company.
Text · Beginner · about 5 minUse cases
Practical patterns with ready-to-run examples. Open any of them in the Playground, then adapt the questions to your data.
Tag financial news for trading signals
Turn each headline or news item into structured fields: is it about this ticker, what kind of event, which direction, and how material, ready for a screener or a model.
See the example →AI safetyGuardrails for LLM apps
Screen every prompt (and reply) for jailbreaks, harmful requests and self-harm in ~100 ms, then pass, review, block or support.
See the example →Customer supportSupport ticket triage and routing
Ask every routing question in one call (team, urgency, refund, frustration), then route in code and send uncertain tickets to a human.
See the example →Search and RAGFilter retrieved passages before your LLM answers
Check each retrieved passage for relevance, real answer evidence, conflicts and prompt injection, so only usable text reaches the answering model.
See the example →VerificationVerify claims and citations against a source
For each claim an LLM (or a writer) makes, check whether the cited source supports it, contradicts it, or says nothing, and only auto-accept confident supports.
See the example →