Decisions & products

TypeScript SDK methods for decision memory (log, list, outcome, execute, delete) and the high-level product endpoints for decision, optimize, retrieve, forecast, and agent runs.

These AlgentaClient methods cover two families: decision memory — persisting the decisions you make and the outcomes you observe (logDecision, listDecisions, getDecision, recordOutcome, executeDecision, deleteDecision) — and the high-level product endpoints that wrap common workflows (productDecision, productOptimize, productForecast, productRetrieve, productAgentRun). Method names are camelCase; bodies stay snake_case.

Decision memory methods

Log a decision when you make it, then record the real outcome later so the engine can learn. logDecision requires only chosen_action; recordOutcome requires actual_outcome.

import { AlgentaClient } from "algenta-sdk";

const client = new AlgentaClient({ apiKey: process.env.ALGENTA_API_KEY });

// 1. Record the decision you made.
const logged = await client.logDecision({
  chosen_action: "launch_now",
  context: "Q3 product launch",
  options_considered: ["launch_now", "wait_a_quarter"],
  expected_value: 142000,
  confidence: 0.81,
  rationale: "Upside outweighs the delay risk.",
});

// 2. Later, attach the observed outcome.
const withOutcome = await client.recordOutcome(logged.id, {
  actual_outcome: 137500,
  outcome_notes: "Slightly under forecast; supply delay.",
});
console.log(withOutcome.id, withOutcome.actual_outcome);

// 3. Browse decisions that already have an outcome.
const page = await client.listDecisions({ page: 1, limit: 25, with_outcome_only: true });
console.log(page);

Method	HTTP	Notes
logDecision(request)	POST /v1/decisions	request is a LogDecisionRequest; only chosen_action is required.
listDecisions(options?)	GET /v1/decisions	Paginate with page, limit, page_size; with_outcome_only filters to decisions with a recorded outcome.
getDecision(decisionId)	GET /v1/decisions/{id}	Fetch one logged decision.
recordOutcome(decisionId, request)	PATCH /v1/decisions/{id}/outcome	request is a RecordOutcomeRequest (actual_outcome, outcome_notes).
executeDecision(decisionId, request)	POST /v1/decisions/{id}/execute	request is an ExecuteDecisionRequest; webhook_url is required.
deleteDecision(decisionId)	DELETE /v1/decisions/{id}	Removes a logged decision.

executeDecision posts to a webhook_url you supply. Set timeout_seconds for slow endpoints, and only pass force or override_safety when you intend to bypass the engine's safety checks.

Product methods

The product endpoints are self-contained wrappers over common decision workflows — you pass a domain-shaped request and get a domain-shaped response, no scenario wiring required.

// A decision from labeled inputs.
const decision = await client.productDecision({
  inputs: [
    { name: "revenue", value: 140000, low: 80000, high: 200000, unit: "USD" },
    { name: "cost", value: 65000, low: 40000, high: 90000, unit: "USD" },
  ],
  objective: "maximize_net_value",
  risk_tolerance: "balanced",
  scenarios: 10000,
});
console.log(decision);

// A forecast from a metric history.
const forecast = await client.productForecast({
  metric: "weekly_active_users",
  history: [1200, 1315, 1402, 1490, 1555],
  horizon: 4,
  seasonality: true,
  confidence_level: 0.9,
});
console.log(forecast);

Method	HTTP	Notes
productDecision(request)	POST /v1/decision	inputs is an array of { name, value, low?, high?, unit? }; add objective, risk_tolerance, scenarios.
productOptimize(request)	POST /v1/optimize	objective plus variables ({ name, min, max, step? }); optional constraints, iterations.
productForecast(request)	POST /v1/forecast	metric and a numeric history; optional horizon, seasonality, confidence_level.
productRetrieve(request)	POST /v1/retrieve	query plus inline documents or a collection_id; optional top_k, rerank.
productAgentRun(request)	POST /v1/agent/run	task to run through the tool loop; optional context, tools, max_steps, output_format.

Optimize over bounded variables

productOptimize searches a bounded variable space for the value of the objective. Each variable declares min/max (and optional step); constraints are expressions with an optional required flag.

const optimized = await client.productOptimize({
  objective: "maximize_net_value",
  variables: [
    { name: "unit_price", min: 9, max: 19, step: 1, unit: "USD" },
    { name: "ad_spend", min: 0, max: 50000, step: 5000 },
  ],
  constraints: [{ expression: "ad_spend <= 0.2 * revenue", required: true }],
  iterations: 500,
});
console.log(optimized);

Next steps

Simulate and query

Run Monte Carlo decisions and governed data queries.

Use the LLM surface

Chat, embeddings, tokenization, and model discovery.

Set up the client

Install, configure, and handle errors.