Docs · Aperture

Quickstart

Create a key in the console, hand your agent llms.txt, and make your first call in under 30 seconds.

export APERTURE_API_KEY="sk_live_..."

curl https://api.aperture.network/v1/chat/completions \
  -H "Authorization: Bearer $APERTURE_API_KEY" \
  -d '{"model":"claude-opus-4-8",
       "messages":[{"role":"user","content":"Research 20 Series-B SaaS VPs of Sales."}]}'

Authentication

Every request authenticates with a single APERTURE_API_KEY sent as a Bearer token. One key reaches every model, API, and skill — there are no per-provider credentials to manage.

Scope each key with its own spend cap, rate limit, and alert thresholds so autonomous agents stay within budget. Keys are created and rotated in the console.

Model Gateway

Call 1,000+ LLMs through one OpenAI-compatible endpoint. Switch models by changing the model field — no re-integration, no new SDK.

Models from Anthropic, OpenAI, Google, DeepSeek, Alibaba, and others are addressable by name, e.g. claude-opus-4-8, gpt-5.5, or gemini-3.5-flash.

API Marketplace

Beyond models, call third-party APIs for search, data, comms, and finance through the same key and the same billing — no separate contracts or onboarding.

Browse the full catalog on the catalog page.

Skills

Skills are packaged, multi-step workflows your agent invokes as a single call — cold-outreach, deep-research, competitor-watch, and more. Each one orchestrates several models and APIs under the hood.

Machine Payments

Every paid endpoint settles per call over Circle Nanopayments, the Machine Payment Protocol (MPP), or the HTTP 402 (x402) flow.

On a 402 Payment Required response, your agent negotiates the quoted micropayment and retries with the receipt attached — no accounts, no invoices, no human in the loop.

Foundry

Foundry runs cloud-hosted, production-ready agent instances that combine the gateway, skills, and payments into one autonomous workflow — with monitoring, guardrails, and payment-native billing built in.

SDK & llms.txt

Official TypeScript and Python SDKs are on the way. Until then every endpoint is plain HTTP and speaks the x402 flow natively — or hand your agent llms.txt and it self-configures.

Developer documentation