Documentation

Get your first guarded LLM call working in under 5 minutes. Create an API key, set the base URL, and make a request.

Step-by-step integration guides for LangChain, Claude Code, CrewAI, OpenAI Agents SDK, Cursor, Aider, and more.

Complete OpenAPI specification for all endpoints — completions, keys, policies, batch jobs, and Anoman-native extensions.

Official Python and TypeScript SDKs with transparent batch polling, streaming support, and type-safe responses.

How injection detection, PII masking, content moderation, and policy enforcement work — and how to configure them.

Traces, live event stream, anomaly detection, and provider health monitoring. Every request is fully visible.

How to route by region and verify where your data is processed. Every model declares its data-processing region; cross-border processing is opt-in.

API key format, Authorization header, key scoping, rotation, and secret-storage best practices.

Every HTTP status and error.code the gateway returns, what causes it, and how your client should handle each one.

Server-Sent Events for token-by-token output. Wire format, mid-stream error handling, cancellation, multi-language samples.

Per-customer token caps by model tier class plus RPM/TPM burst guards and a concurrent inflight cap. Response headers, 402/429 handling, token packs for planned spikes.

Asynchronous events for batch completions, anomaly detection, balance thresholds. HMAC signature verification, at-least-once delivery, idempotency patterns.

Per-endpoint detail pages. Chat completions, batch jobs, embeddings, models — full parameter tables, response shape, headers, and code samples.

How the system works behind the API. Weighted tokens, guardrails pipeline order, caching layers — the model you need to budget + debug effectively.

Two-line migration from raw OpenAI or Anthropic SDK. Keep all your code; gain guardrails, observability, and the full multi-provider catalog.

End-to-end working code for the patterns customers ask about most. Code review bot, streaming chat UI, batch document analysis.

The four tiers, weighted-token billing, annual billing, and IDR payment via DOKU.

Monthly weighted-token allowance, per-model-class token caps (402), overage vs block, and resets.

Discover models, tier classes, capability flags, offered = active + public + reachable, sovereign routing.

Buy token packs to temporarily raise a model tier class's token cap when you plan a spike.

Anoman Design — POST /v1/images/generations, Flux models, enhance + style presets, and the studio.

copilot.anoman.io branded chat — entitled models, history, streaming, SSO + email-OTP MFA.

Scan model responses (egress) for secrets and PII. Modes off / monitor / redact / block.

Detect jailbreak / DAN attempts and indirect injection from tool results. Modes off / monitor / block.

Presidio detection incl. NIK / NRIC. Redact, tokenize, synthetic, or block — with de-anonymization.

Govern MCP tool calls — register servers, per-tool allow / deny / require_approval, scopes, audit.

Bundle guardrail config per API key. Per-key overrides, priority order, and a master toggle.

Organization accounts and roles — create, switch, convert, and transfer ownership.

Invite staff by email — 7-day token, accept flow, revoke. Roles admin / admin_read_only / member.

Entitlement bundles — allowed tier classes, per-member weighted caps, and a member-domain allowlist.

Keyless endusers via Copilot — single + bulk invite, suspend, reactivate, reset, offboard.

Refer & Earn — 20% bill credit per side on a referee's first 3 settled payments.

Every request recorded — tokens, cost, latency, provider, guardrails, cache, routing, sessions.

Weighted tokens used vs allowance, spend by model / region / agent, savings, and batch economics.

Real-time SSE stream of requests, blocks, anomalies, batch jobs, and per-provider health.

Z-score / Welford rolling detector plus per-customer Isolation Forest ML. List + acknowledge.

Append-only audit_events (SOC 2 evidence) — cursor-paginated list plus NDJSON export.

Two stacking layers — semantic response cache plus provider prompt cache. Weighted-token discounts.

Route non-interactive work through batch for ~50% lower cost. 202 → poll → 200 with savings.

Enterprise SAML single sign-on via saml-jackson — any SAML 2.0 IdP. Admin CRUD, tier-gated.

Opt-in preview response-groundedness signal — CPU, async, warn-only. Coarse band, never blocks.