A language for continuous decisions
Write the computation in SQL, Python, or JavaScript. Replay it on history, inspect every decision, then keep the same stateful program running as new data arrives.
Loading synchronized BTC and ETH market history…
01 / STREAMING BY CONSTRUCTION
RtBot SQL compiles into a stateful operator graph. Each (time, value) tuple advances that graph incrementally, and an output appears only when the program has something to say.
02 / EVENT TIME, NOT ARRIVAL TIME
Every value carries its own timestamp. RtBot queues streams independently and evaluates a multi-stream expression only when event times match. One stream may lead, lag, or arrive in bursts; as long as each stream stays ordered, arrival timing does not change the causal result.
| TIME | A | B | A + B |
|---|
Result appears only when A(t) and B(t) both exist.
03 / REPRODUCIBLE BY CONTRACT
Run the same timestamped inputs through the same program and initial state. Packets may pause, burst, or interleave differently in physical time; the event-time output trace remains identical.
CONTRACT Each input preserves event-time order. Only the cross-stream wall-clock schedule changes.
04 / SIGNALS, NOT ROWS
When the problem is one causal equation over several ordered signals, RtBot keeps the domain model intact. Time is not another field to reconcile. It is the coordinate system on which the program runs.
active_power(t) = voltage(t) × current(t) × power_factor(t)ONE EQUATION · THREE LIVE SIGNALSSELECT
v.event_time,
v.value * i.value * pf.value AS active_power
FROM voltage AS v
JOIN current AS i
ON i.event_time = v.event_time
JOIN power_factor AS pf
ON pf.event_time = v.event_time;Join voltage to currentState that their event-time columns must match.
Join the power factorRepeat the temporal relationship for the third signal.
Then evaluate the equationThe physical calculation comes after row matching.
SELECT
v.value * i.value * pf.value AS active_power
FROM voltage AS v,
current AS i,
power_factor AS pf;List the three signalsEach input is already an ordered signal x(t).
Time alignment is automaticThe three signal fronts meet at event time t.
Emit active_power(t)The result appears when the equation is evaluable.
BEFORETranslate one equation into a chain of temporal joins.
WITH RTBOTKeep the equation as the executable program.
05 / ONE PROGRAM, TWO MODES
RtBot removes the production rewrite from the development cycle. Historical and live events advance the same compiled graph, with the same state transitions and output semantics.
Every translation is another place for the algorithm and the production system to disagree.
Replay stops. Live input starts. The program does not change.
06 / CONCRETE PROOF
NASA IMS Dataset 1 provides a public 34-day failure history and published signatures against which every step can be inspected.
CREATE MATERIALIZED VIEW rms_alert AS
SELECT device_id, channel_id, rms_value,
MOVING_AVERAGE(rms_value, 48) AS baseline_rms
FROM rms_trend
WHERE rms_value > MOVING_AVERAGE(rms_value, 48) * 2.5;07 / WHERE IT FITS
RtBot earns its place when a decision depends on history, context, several signals, and exact timing—not merely the latest row.
Know when equipment, infrastructure, or a process begins to degrade—not merely when one reading crosses a line.
TYPICAL OUTPUT · DEGRADATION STATETurn fast-moving activity into explainable exceptions, flags, and escalations while preserving the evidence behind them.
TYPICAL OUTPUT · FLAG · ESCALATIONCombine several signals, persistence rules, and operating context before interrupting a line or involving an operator.
TYPICAL OUTPUT · HOLD · INTERVENTIONContinuously evaluate service, business, and control signals and emit the small number of changes downstream systems need.
TYPICAL OUTPUT · STATE CHANGE · OPPORTUNITY08 / START FREE, STAY PORTABLE
Use RtBot directly inside applications you own. Explore and backtest in Python, make the computation tangible in JavaScript, then embed the same deterministic graph in a service.
Replay data frames, tune parameters, inspect intermediate outputs, and embed the native runtime in numerical workflows.
Open the Python guide →Run the same engine in browsers, workers, and JavaScript applications—like the live computation at the top of this page.
Open the JavaScript guide →Run the same graph inside native or JVM services and keep execution close to the systems that already own the data.
Inspect the runtime source ↗09 / FROM PROGRAM TO OPERATIONS
A Helm chart installs the data plane in your Kubernetes cluster. The Kernel Controller manages the complete lifecycle of the self-hosted deployment.
Manages the kernel lifecycle
WHEN TO CALL US
We help technical teams move validated algorithms from batch or micro-batch pipelines into continuous, self-hosted production systems. Bring one existing computation and the constraints around it.
Discuss your streaming migration FREE TO START · COMMERCIAL IMPLEMENTATION AND SUPPORTTALK TO US IF YOU NEED TO:
Convert batch or micro-batch computations into continuous streaming programs.
Rewrite full-window jobs as incremental, rolling state updates.
Validate and tune against historical data, then run the same program live.
Process high-frequency, time-stamped data at sustained throughput.
Combine asynchronous streams by event time without custom synchronization.
Deploy and operate the complete data plane inside your own infrastructure.