SuperAPI AI Directory

decide

Deterministic refund eligibility checks for subscriptions (Adobe, Spotify, etc.), returning ALLOWED, DENIED or UNKNOWN from a stateless rules engine.

Open website

About this tool

decide.fyi is a deterministic decision API engine whose primary purpose is to perform provenance-supported, fail-closed policy checks in workflow applications, MCP notarization remote services, decision memo data packets, and execution gating, covering refunds, cancellations, returns, and trials. It allows an agent's proposed action basis to be evaluated against a deterministic rulebook before agent tools gain permission to change state, and callers receive a replayable decision record and are routed to yes, no, or review while retaining credentials and execution control.

In terms of main capabilities, the production core is hybrid_declarative_rulebook_with_trusted_adapters, which supports direct declarative rulebooks and also allows registered first-party trusted adapters to provide bounded facts, but rejects customer-executable rulebooks. In both binding modes, Rulebook v1 is the only binding adjudication selector. mode: "rulebook" does not call an LLM; it validates the request rulebook against the published JSON Schema, hashes the rulebook, evaluates bounded conditions, and returns yes, no, or review, along with the applied adjudication, action, reason codes, matched rules, and evaluator_version. The response also includes rulebook_contract (the enforced schema URL/hash), runtime_binding (direct or trusted adapter binding mode), input_hash, and the rulebook_attestation_v1 bundle hash. Production deployments can sign that bundle hash with the rulebook_attestation_signature_v1 Ed25519 envelope, with verification keys published at /.well-known/rulebook-attestation-keys.json. Setting DECIDE_RULEBOOK_ATTESTATION_SIGNATURE_REQUIRED=true can fail closed in production instead of returning an unsigned rulebook decision.

In terms of usage, the public @decide-fyi/sdk client covers Decision API calls, Decision Record and Decision Packet verification, rulebook consistency, replay, execution receipts, and outcome reporting. The SDK source is in the sdk/ directoryand is licensed under Apache-2.0. It provides four one-click-install Decide Policy Notaries servers, and existing professional endpoints remain supported compatibly. The stable MCP remote service is stateless, requires no authentication, and allows 100 requests/minute, and can be connected to Claude Desktop, Windsurf, or other clients via MCP. Local development checks include starting a local development server. The protected Zendesk reference workflow route returns the consultation category, policy outcome, and recommended Zendesk action shape, but does not write to Zendesk, does not authorize execution, and does not create a binding production decision record; every response sets workflow_contract.execution_allowed: false. Production workflow requests require a server-to-server Bearer token.

The four core endpoints and MCP tools are respectively: POST https://refund.decide.fyi/api/v1/refund/eligibility(MCP tool refund_eligibility), evaluates refunds, with inputs vendor, days_since_purchase, region, plan, and conditional qualifying_conditions_met, and codes including WITHIN_WINDOW, OUTSIDE_WINDOW, NO_REFUNDS, MISSING_REQUIRED_CONTEXT, UNSUPPORTED_VENDOR; POST https://cancel.decide.fyi/api/v1/cancel/penalty(MCP tool cancellation_penalty), checks cancellation penalties, with inputs vendor, region, plan, and conditional billing_cadence, and codes including NO_PENALTY, EARLY_TERMINATION_FEE, CONTRACT_LOCKED, MISSING_REQUIRED_CONTEXT, UNSUPPORTED_VENDOR; Adobe cancellation is review-only mode; POST https://return.decide.fyi/api/v1/return/eligibility(MCP tool return_eligibility), evaluates return eligibility, with codes including FULL_RETURN, PRORATED_RETURN, CREDIT_RETURN, OUTSIDE_WINDOW, NO_RETURNS, MISSING_REQUIRED_CONTEXT, UNSUPPORTED_VENDOR; POST https://trial.decide.fyi/api/v1/trial/terms(MCP Tool trial_terms), evaluate availability and terms based on real-time offer observation, with inputs vendor, region, plan, offer_confirmed, and the confirmed observed_trial_days, observed_card_required, observed_auto_converts; codes include AUTO_CONVERTS, NO_AUTO_CONVERT, TRIAL_NOT_AVAILABLE, MISSING_REQUIRED_CONTEXT, UNSUPPORTED_VENDOR.

Applicable scenarios include eligibility verification for subscription refunds, cancellations, returns, and trial terms, as well as reference applications such as AI agent execution gating, workflow readiness binding, execution gates, and decision memo preparation. Regarding limitations, policy coverage defaults to US individual plans, Typeform is limited to cancellation (Basic, direct, self-service platform subscription coverage), known vendor IDs do not mean all scopes can be automated, and automation_safe, policy_decision_mode, and policy_evidence in each response must be checked. WeightWatchers refund, cancellation, and return sources have been retired; their history is available but they are no longer actively fetched or used for new automated decisions. Missing or outdated evidence will be routed to review. Legacy single, multi, and runtime requests can still be used for AI-assisted exploration, but carry authority: "advisory_only" and production_verdict: false, and are not binding production verdicts. Trial results require real-time offer observation. Approval-related policies remain UNKNOWN even if the caller sets qualifying_conditions_met. Evidence fields are set only when source-specific conditions are actually verified; a time window by itself does not constitute proof.