# Secton API

> OpenAI-compatible inference API from Secton (secton.org). Two published operations: chat
> completions (with optional streaming) and model listing. Authentication is a static API key
> issued in the Secton Console and sent in the `Authorization` header.

GENERATED by the API Evangelist enrichment pipeline on 2026-08-16. Secton does not publish an
llms.txt of its own — /llms.txt returns 404 on secton.org, console.secton.org and
platform.secton.org, and the docs host named in earlier notes (dev.secton.org) does not resolve in
DNS. Everything below is sourced from Secton's own published OpenAPI, legal pages, blog and npm
package, each cited inline.

## Read this first — lifecycle status is unresolved

On 2025-11-18 Secton announced it was "permanently shutting down Copilot, Secton Platform, and
Secton API", with the API operational only until 2025-12-19 and customer data deleted after that
date (https://secton.org/blog/a-new-chapter).

As of 2026-08-16 the API host still answers, still issues auth challenges pointing at the key
console, still serves its OpenAPI, and its Console Terms of Service were re-issued effective
2026-08-06 with sections governing API keys, usage limits and billing. Secton has not published a
retraction of the shutdown notice. Confirm the API's status with Secton before building on it.

## Base URL

https://api.secton.org

## Authentication

- Scheme: API key in the `Authorization` request header (OpenAPI scheme name `ApiKeyAuth`).
- Both `Authorization: <key>` and `Authorization: Bearer <key>` were accepted when probed.
- Keys are issued at https://console.secton.org/api (account required).
- SDK environment variable: `SECTON_API_KEY`.
- No OAuth, no OIDC, no scopes.

## Operations

### POST /v1/chat/completions — Create a chat completion
- operationId: `post-external-api-v1-chat-completions`
- Request: `model` (string, required), `messages` (array of `{role: system|user|assistant,
  content: string}`, required, min 1), `temperature` (number, default 0.7), `stream` (boolean,
  default false), `max_tokens` (integer, 0–4096).
- Response 200: `{id, object: "chat.completion", created, model, choices[{index, message{role,
  content}, finish_reason}], usage{prompt_tokens, completion_tokens, total_tokens}}`.
- Streaming: with `stream: true` the response is a sequence of `chat.completion.chunk` frames
  (`choices[].delta{role?, content?}`, terminal when `finish_reason` is non-null). NOTE: the
  published spec expresses this through an unresolvable `$ref` and it is not usable as-is — see
  the corrected form in overlays/secton-api-chat-api-overlay.yaml.
- NOT idempotent. No idempotency key exists; a retry produces a second billable generation.

### GET /v1/models — List available models
- operationId: `get-external-api-v1-models`
- No parameters. Response 200: `{object: "list", data: [{id, object: "model"}]}`.
- Unpaginated and unfiltered. Visibility is scoped to the calling key.
- Known model identifiers published by Secton: `throb-v1`, `copilot-zero`. Not exhaustive.

## Errors

Envelope is `{"error": "<human-readable string>"}` with `application/json`. Not RFC 9457, and there
is no machine-readable error code — branch on the HTTP status.

- 401 `API key is missing from bearer. Get your API key at https://console.secton.org/api`
- 401 `Invalid or expired API key`
- 429 rate limited — inferred from the SDK's `RateLimitError.retryAfter`, not confirmed.

Neither operation declares any 4xx/5xx response in the published OpenAPI.

WARNING for crawlers and agents: api.secton.org returns **HTTP 200** for unrouted paths, with the
body `{"message":"The <path> endpoint doesn't exist!"}`. A 200 from this host is not evidence that
a resource exists. Check the body.

## Rate limits and pricing

Neither is published. Console Terms §13 enumerates limit categories (request, rate, token,
throughput, storage, concurrency, workspace, organization, model-specific) but states values live
"through the Console" — i.e. behind login. Console Terms §14 says services "may be provided without
charge" and that paid tiers "may" be introduced; there is no pricing page (secton.org/pricing 404s).
Token usage is metered and returned on every completion; the SDK reads a credit balance
(`creditsAvailable`/`creditsTotal`) from an endpoint that appears in no published spec.

## Client libraries

- npm `secton` (TypeScript/JavaScript), v1.0.2 published 2025-07-04, MIT, source at
  https://gitlab.com/wearesecton/secton-sdk. **Every published version carries an npm deprecation
  flag.** No Python, Go, Ruby, Rust, PHP, Java or .NET client exists.

## Agent surfaces

- OpenAPI: https://console.secton.org/openapi.json (OpenAPI 3.0.0, 2 operations)
- MCP server: none.
- A2A agent card: none — /.well-known/agent-card.json and /.well-known/agent.json miss on every host.
- Webhooks / events / AsyncAPI: none. Streaming is in-band on the chat request.
- /.well-known/ documents: none on any host.

## Links

- Company: https://secton.org
- Console and API keys: https://console.secton.org/api
- OpenAPI: https://console.secton.org/openapi.json
- Status: https://status.secton.org
- Security / responsible disclosure: https://secton.org/security
- Blog: https://secton.org/blog
- Console Terms of Service: https://secton.org/legal/console-terms
- Privacy Policy: https://secton.org/legal/privacy
- Copilot Usage Policy: https://secton.org/legal/copilot-usage
- Contact: https://secton.org/contact
- GitHub: https://github.com/sectoncorp
- SDK source: https://gitlab.com/wearesecton/secton-sdk