Guides / Get started

Quick Start

Go from zero to your first routed model call in under five minutes.

Create an API key Explore endpoints

API base URLhttps://origamiaai.vercel.app/v1

OpenAI compatibleAnthropic MessagesSSE streamingHard budget controls

Get your API key

Open the Origami Console and sign in — or create a workspace.
Go to API keys and create a key. Keys are scoped per project.
Copy the key — it is shown once, prefixed ori_.

API Keys Management

Create, scope, and rotate virtual keys in the console.

Billing & Usage

Review the wallet balance and track per-key spend in real time.

Choose a model

Every model on Origami is reachable through the same endpoint — just change the model field. Here are a few currently available text models:

Anthropic Claude Fable 5

anthropic/claude-fable-5

Claude Fable 5 is our most capable generally available model for autonomous knowledge work and coding. It is a Mythos-class model with robust safeguards that can handle long-running, complex, and asynchronous tasks.

Anthropic Claude Haiku 4.5

anthropic/claude-haiku-4-5

Fastest Claude. Low latency and cost.

Anthropic Claude Opus 4.7

anthropic/claude-opus-4-7

Claude Opus 4.7 is Anthropic’s high-tier model designed for superior autonomy and complex reasoning. It excels at long-horizon software engineering, allowing users to delegate difficult, multi-step coding tasks with significantly less human supervision. The model features tripled vision resolution (up to 2,576 pixels), enabling it to analyze dense screenshots and extract precise data from complex diagrams.

Anthropic Claude Opus 4.8

anthropic/claude-opus-4-8

Claude Opus 4.8 is Anthropic's flagship model featuring sharper judgment, greater honesty about its own progress, and the ability to work independently for longer than its predecessors. It leads on agentic coding (SWE-Bench Pro: 69.2%), multidisciplinary reasoning, and agentic computer use. Fast mode runs at 2.5x speed at 3x lower cost than previous fast mode versions.

Pick a calling method

Use raw HTTP, an official SDK, or keep your existing OpenAI client — Origami is fully compatible.

HTTP API

Standard RESTful endpoint, compatible with every language and stack.

Python SDK

Official SDK with full type hints, streaming, and async support.

Node.js SDK

Typed JavaScript client for server and edge runtimes.

OpenAI SDK

Drop-in OpenAI compatibility — change two lines and migrate.

Make your first call

With your key and a model selected, send your first request:

1curl "https://origamiaai.vercel.app/v1/chat/completions" \
2  -H "Authorization: Bearer $ORIGAMI_API_KEY" \
3  -H "Content-Type: application/json" \
4  -d '{
5    "model": "YOUR_PUBLIC_MODEL",
6    "messages": [{"role": "user", "content": "Hello from Origami"}]
7  }'

Responses follow the OpenAI schema, so existing parsing code keeps working. Streaming, tool calls, and JSON mode are supported across all chat models.

What's next

Set hard budgets

Cap spend per key, team, or model before traffic scales.

Smart routing

Let the gateway pick the fastest, cheapest healthy provider.

API Reference

Full parameter docs for every endpoint.

Access

Authentication and key policy

Every public API request is authorized against the virtual key and its active policy.

Bearer authentication

Use Authorization: Bearer $ORIGAMI_API_KEY. The Messages endpoint also accepts x-api-key for Anthropic SDK compatibility.

Server-enforced policy

Activation, expiry, rotation grace, revocation, model rules, modality permissions, CIDR restrictions, budgets, and quotas are evaluated before routing.

List available models

1curl "https://origamiaai.vercel.app/v1/models" \
2  -H "Authorization: Bearer $ORIGAMI_API_KEY"

A key only sees active public models permitted by its account and access policy.

Platform

How the API system works

Each request follows a consistent lifecycle from authorization through usage recording.

01Authorize

Validate the virtual key and confirm the requested model and capability are available.

02Apply limits

Check the request against the key's active usage, budget, and concurrency limits.

03Process

Send the request to an available model service that supports the required capabilities.

04Record

Return the response and update the request's usage and cost record.

Platform

Models, routing, and capabilities

Public model

The stable ID sent by customers and returned from GET /v1/models.

Capabilities

Features available to the model, such as tools, vision, reasoning, media, audio, files, and batch processing.

Availability

Only active models permitted by the key are returned and accepted by the API.

Compatibility

Use the OpenAI-compatible or Anthropic-compatible API surface supported by your client.

Origami maintains compatible request, response, usage, and streaming behavior across its supported client formats.

Platform

Usage, budgets, and accounting

Origami checks limits before a request and records its usage and final cost.

1. Check

The request is checked against the available balance and every active limit.

2. Measure

Text, cache, media, audio, file, and request usage is recorded when applicable.

3. Report

The dashboard shows the request status, measured usage, and resulting cost.

If exact usage is unavailable, the cost may be marked as estimated. View request-level details in Usage.

Platform

Limits, idempotency, and reliability

Rate limits

Effective RPM, TPM, daily/monthly/rolling token and cost, plus normal and streaming concurrency limits are enforced atomically.

Idempotency

Send Idempotency-Key with 1–255 printable ASCII characters. Keys are scoped to virtual key and endpoint.

Replay

Completed non-stream responses may be replayed. Completed streams are not replayable, but duplicate execution remains blocked.

Request identity

Store the returned X-Request-ID when contacting an administrator or investigating a failure.

Rate-limit response headers

1RateLimit-Limit: 60
2RateLimit-Remaining: 42
3RateLimit-Reset: 18
4RateLimit-Policy: 60;w=60;comment="key"
5X-Request-ID: req_...