# Ward > Ward is Mandate Labs' decision model for fraud, AML, compliance and risk: send the input, get APPROVE / STEP_UP / DECLINE with the score, signals and typologies your code can act on. ## Introduction - [Introduction](https://ward.mandatelabs.ai/): What Ward is, and how to send it an input and get a decision your code can act on. - [Quick start](https://ward.mandatelabs.ai/quickstart): Get an API key and make your first Ward decision in about five minutes. - [Example use cases](https://ward.mandatelabs.ai/use-cases): Industry workflows that become a Ward call and a branch in your code. ## Concepts - [How Ward decides](https://ward.mandatelabs.ai/concepts/decision-model): Ward is a decision model, not a chat model: narrow, typed decisions with published signals, so your code stays in control. - [Input](https://ward.mandatelabs.ai/concepts/input): What to send for each decision type, how the type is inferred, and how much input one call can take. - [The decision envelope](https://ward.mandatelabs.ai/concepts/envelope): Every field of a Ward answer, the same for every decision type. - [Scores and confidence](https://ward.mandatelabs.ai/concepts/scores): What each score means, which scores are probabilities, and how to use them with thresholds. - [Sessions, context window and billing](https://ward.mandatelabs.ai/concepts/sessions-billing): How sessions group calls, the context window per session, and per-token billing from a prepaid balance. - [Subject memory](https://ward.mandatelabs.ai/concepts/subjects): Link decisions about one customer with subject_ref, record consent, and erase a customer's memory. - [Feedback and learning](https://ward.mandatelabs.ai/concepts/feedback): Tell Ward whether a decision was right; feedback tunes thresholds and trains a model for your account that stays advisory until it proves itself. ## Capabilities - [Capabilities](https://ward.mandatelabs.ai/capabilities/): The decision types Ward answers, what each takes and returns, and how they combine. - [Payment](https://ward.mandatelabs.ai/capabilities/payment): Decide a card, bank-transfer, wallet or crypto payment, whether a person or an AI agent initiated it. - [Agent intent](https://ward.mandatelabs.ai/capabilities/intent): Check an AI agent's intended action against its mandate before it acts. - [Agent (KYA)](https://ward.mandatelabs.ai/capabilities/agent): Know Your Agent: verify an AI agent's live trust status, KYA Score and active mandate. - [Screen](https://ward.mandatelabs.ai/capabilities/screen): Screen a person or organization against sanctions, PEP, regulatory enforcement and adverse media, telling namesakes apart. - [Document](https://ward.mandatelabs.ai/capabilities/document): Verify an identity document's machine-readable zone, compare it with declared data, and score signs of editing. - [Statement](https://ward.mandatelabs.ai/capabilities/statement): Read bank statements, card exports or alert files in any layout and score AML and fraud risk across the whole history. - [Account takeover](https://ward.mandatelabs.ai/capabilities/ato): Read an account's logs before a disputed transaction and tell third-party fraud from first-party misuse. - [Credit](https://ward.mandatelabs.ai/capabilities/credit): Read affordability from bank statements: income regularity, obligations, burden, buffers and loan stacking. - [Merchant](https://ward.mandatelabs.ai/capabilities/merchant): Merchant and business risk for onboarding (KYB) and monitoring: lists on the business and owners, WMID, website, rates. - [KYC](https://ward.mandatelabs.ai/capabilities/kyc): Verify an applicant from your own onboarding UI: identity screening, email, phone, ownership codes and the ID document, with reasons you can show. - [Ask](https://ward.mandatelabs.ai/capabilities/ask): Ask questions about a decision's result in plain language; answers are built from the result's own data, never generated. ## Guides - [Files in any format](https://ward.mandatelabs.ai/guides/files): How Ward reads statements and exports in any column layout, and how to confirm a reading once per layout. - [Async jobs and idempotency](https://ward.mandatelabs.ai/guides/async-idempotency): Run large decisions in the background, and make retries safe so nothing is decided or billed twice. - [Evidence pack](https://ward.mandatelabs.ai/guides/evidence): Everything an analyst needs to act on a decision, including a goAML-shaped STR draft. Ward files nothing. - [Thresholds](https://ward.mandatelabs.ai/guides/thresholds): Set your own cut-offs per capability, per request or per account, and use feedback-driven suggestions. ## Patterns - [Patterns](https://ward.mandatelabs.ai/patterns/): Architectural patterns for building on Ward decisions. - [Step-up routing](https://ward.mandatelabs.ai/patterns/step-up-routing): Route STEP_UP to the cheapest friction that resolves it, using signals and score kind. - [Fail-closed handling](https://ward.mandatelabs.ai/patterns/fail-closed): Ward never reports an unchecked list or a failed lookup as clear; handle not_checked explicitly. - [Review queue](https://ward.mandatelabs.ai/patterns/review-queue): Rank the decisions people must review, and close the loop with feedback. ## Cookbooks - [Cookbooks](https://ward.mandatelabs.ai/cookbooks/): End-to-end recipes built on Ward decisions. - [KYC onboarding](https://ward.mandatelabs.ai/cookbooks/kyc-onboarding): A complete onboarding flow on your own UI with Ward KYC, from codes to an account decision. - [AML statement review](https://ward.mandatelabs.ai/cookbooks/aml-statement-review): Analyse a customer's statements, review the flagged movements with Ask, and export the evidence pack. - [Chargeback triage](https://ward.mandatelabs.ai/cookbooks/chargeback-ato): Decide whether a fraud dispute is third-party fraud after an account takeover or first-party misuse. ## Models - [Models](https://ward.mandatelabs.ai/models): Ward releases, aliases, component versions and the changelog; how to pin a model. ## API - [Decide](https://ward.mandatelabs.ai/api/decide): POST /api/v1/decide — one decision for any input, billed on input tokens. - [Ask](https://ward.mandatelabs.ai/api/ask): POST /api/v1/ask — a question about a result, answered from the result's data. - [Jobs](https://ward.mandatelabs.ai/api/jobs): GET /api/v1/jobs/{id} — the status and answer of an async decision. - [Feedback](https://ward.mandatelabs.ai/api/feedback): POST /api/v1/decide/{id}/feedback — tell Ward whether a decision was right. - [List models](https://ward.mandatelabs.ai/api/models): GET /api/v1/models — the releases, component versions, aliases and changelog. - [Usage and balance](https://ward.mandatelabs.ai/api/usage): GET /api/v1/billing/balance and /usage — prepaid balance and token consumption. - [Errors and limits](https://ward.mandatelabs.ai/api/errors): The error format, every status code Ward returns, rate limits and retry guidance. ## Client SDKs - [Client SDKs](https://ward.mandatelabs.ai/sdks/): Python and JavaScript clients with typed decisions, retries, idempotency and helpers for files, KYC and jobs. - [Python SDK](https://ward.mandatelabs.ai/sdks/python): Install mandate-ward and make sync or async Ward calls from Python 3.9+. - [JavaScript SDK](https://ward.mandatelabs.ai/sdks/javascript): Install @mandatelabs/ward and call Ward from Node 18+, Deno, Bun or edge runtimes, with TypeScript types.