Owctopus
Log in

Decision optimization for AI apps

Stop paying AI to make the same decision twice.

Owctopus is a decision optimization layer for AI-powered applications. It learns validated, reusable conditions from recurring decisions and executes matching requests without another model call.

AI teaches the system. Conditions run the system.

  • Lower inference costs
  • Faster decisions
  • Traceable outcomes

Integrate with your existing AI provider. Keep your own API key. Start with one decision and measure the savings.

Start buildingRead the documentation
strategy conditionrecoveryno AI call

Your application sends

await owctopus.decide({
  decision: "training_recommendation",
  context: {
    sleep_hours: 5.4,
    sleep_quality: "poor",
    resting_heart_rate: 78,
    heart_rate_baseline: 65,
    stress_level: "high",
    workout_yesterday: "heavy",
    workout_days_last_7: 5,
    goal: "fat_loss",
  },
  options: [
    { id: "hard_workout" },
    { id: "moderate_workout" },
    { id: "light_workout" },
    { id: "recovery" },
  ],
});

Owctopus returns

{
  "option": "recovery",
  "confidence": 0.94,
  "strategy": "condition",
  "reasons": [
    "Poor sleep, high stress, and heavy recent training."
  ],
  "nextAction": "prioritize_recovery",
  "executionId": "ex_1842"
}

The call

You send the situation. Owctopus returns a decision you can use.

Your application owns the facts and the options. Owctopus owns whether a learned condition can answer before AI is asked.

You provide

  • decisionWhat should happen — e.g. routing, eligibility, recommendation.
  • contextThe facts for this moment that your product already knows.
  • optionsThe choices your application can actually take.

Owctopus returns

  • optionThe choice to use in your product.
  • strategycondition when a learned pattern matched; ai when the model answered.
  • confidenceHow sure Owctopus is about that option.
  • reasonsWhy this choice won — ready to log or show.
  • executionIdTrace handle for this run.

How it works

Condition first. AI when the pattern is missing.

Owctopus does not hard-code dozens of if/else branches. It learns reusable patterns, validates them, then runs matching requests deterministically.

A learned condition · example

Poor sleep, heavy load, elevated strain → recovery.

  • If sleep hours are under 6
  • and sleep quality is poor
  • and recent training load is high
  • and stress is high or heart rate is elevated
  • then recommend recovery

Unrelated fields stay out of the rule. Conditions are scoped by decision and must be approved before they run alone.

Learned condition

{
  "decision": "recovery",
  "condition": {
    "sleep_hours": "< 6",
    "sleep_quality": "poor",
    "recent_training_load": "high",
    "stress_or_elevated_hr": true
  },
  "source": "learned_condition",
  "confidence": 0.94,
  "observations": 183
}

  1. Evaluate conditions

    Match validated patterns for this decision and tenant.

  2. Return when matched

    strategy: condition. No model call.

  3. Call AI when unknown

    Your provider answers among your options.

  4. Learn candidates

    Repeated drivers become candidate conditions.

  5. Activate after proof

    Test, validate, approve — then the condition can run alone.

  1. CandidateEnough evidence to propose it.
  2. ValidateReplay against past outcomes.
  3. ApproveIt has earned the right to run.
  4. ActiveMatching requests skip the model.
  5. RollbackDisable or expire a bad rule immediately.

Validation is core: unvalidated rules do not activate automatically. See how learning and approval work.

Illustrative ROI

Savings only matter when your numbers work.

Customers pay Owctopus per decision and still pay their AI provider for unknown decisions. Use your model cost and expected condition hit rate — not ours.

AI calls avoided

88.4%

442k of 500k answered by conditions in the illustrative Growth example.

Est. AI cost avoided

$177

At OpenAI GPT-4o mini published rates (~1200 in / 300 out tokens per call).

Owctopus plan cost

$59

Growth tier for that volume — net savings depend on your model price and hit rate.

Illustrative net monthly benefit

$118+

Avoided AI cost minus Owctopus plan — before engineering cost. Not a measured benchmark.

See pricing & estimate ROIRead the docs

Start in minutes

Install it, send one decision, measure condition vs AI.

Keep your AI provider key. Put the Owctopus key on your server. The first call is the whole integration.

  1. 01

    Install

    TypeScript, Python, Go, or HTTP.

  2. 02

    Configure

    OWCTOPUS_API_KEY on your server. Keep your AI key BYOK.

  3. 03

    Decide

    One decide() call — then measure condition vs AI.

import { Owctopus } from "@owctopus/sdk";

const owctopus = new Owctopus({
  apiKey: process.env.OWCTOPUS_API_KEY,
});

const result = await owctopus.decide({
  decision: "training_recommendation",
  context: {
    sleep_hours: 5.4,
    sleep_quality: "poor",
    stress_level: "high",
    workout_yesterday: "heavy",
    workout_days_last_7: 5,
  },
  options: [
    { id: "hard_workout" },
    { id: "moderate_workout" },
    { id: "light_workout" },
    { id: "recovery" },
  ],
});

console.log(result.option);

OWCTOPUS_API_KEY=ow_live_xxxxxxxxx

Production defaults to https://api.owctopus.com. Optionally set OWCTOPUS_BASE_URL for local development.

Suitable decisions repeat with stable patterns and measurable cost or latency. Exceptions and high-stakes choices should stay on review or AI.

Start buildingQuickstart

Owctopus

AI teaches the system. Conditions run the system.

Start with one decision. Measure avoided model calls, cost, and latency — then expand to the workloads that repeat.

Start buildingPricing

Owctopus