DEVELOPERS / GETTING STARTED

One API for the web.

Ptyktos speaks the OpenAI chat-completions wire, so your existing client works by swapping one URL. The rest of the surface is where Ptyktos is alone: extraction, document conversion, and research retrieval — every result carrying its provenance.

QUICKSTART

First request in two minutes.

Generate a key. Sign in to the console and open Account → API keys → Generate. Keys are issued per account and verify on every request.
Store it like a credential — a secrets manager, not a dotfile:
```
export PTYKTOS_API_KEY="your key goes here"
```

Call it. OpenAI-compatible, so your existing client only needs the base URL:

curl -X POST https://api.ptyktos.com/v1/chat/completions \
  -H "Authorization: Bearer $PTYKTOS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"messages": [{"role": "user", "content": "Hello"}]}'

Then use what nobody else serves. Extract a page:

curl -X POST https://api.ptyktos.com/v1/web/scrape \
  -H "Authorization: Bearer $PTYKTOS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com/article"}'

SURFACE

The endpoints.

POST /v1/chat/completions

OpenAI-compatible chat completions, multi-upstream. Drop-in: swap the base_url, keep your client.

POST /v1/web/scrape

URL → clean markdown in a single streaming HTML pass (~150ms/page typical). Provenance attached.

POST /v1/web/crawl

Crawl with the same engine: frontier, politeness, and per-page extraction records.

POST /v1/web/search

Web search with extraction-backed results — search first, then scrape what matters.

POST /v1/doc/convert

DOCX, PPTX, XLSX, ODT, EPUB, RTF, CSV, PDF → structured output. Single-pass, in-process.

POST /v1/arxiv/search

arXiv search with reference pruning and LLM-ready chunking.

POST /v1/youtube/transcript

Transcript extraction from any YouTube URL.

GET /v1/catalog

Hosted MCP tool catalogue — agent-integrable, receipted, governed.

Typed SDKs: npm install ptyktos · pip install ptyktos · cargo add ptyktos. Every operation returns its provenance — the source, the measured result, and the limits of that result.

KEYS AND SAFETY

Keys are credentials. Treat them that way.

Keys verify per request against the credential store — no shared session, no ambient authority. Rotate from the console at any time; old keys stop verifying immediately. Server-side only: a key in a browser bundle is a leaked credential. Metering is per-operation and auditable — your usage log is the same one we reconcile against.

Read how the data path stays on Australian metal ↗