Public search with NLWeb

Query Siglata's website and documentation in natural language, with cited JSON or SSE results.

The https://www.siglata.com/ask endpoint searches Siglata’s public pages and documentation. It returns relevant pages with a title, description, language and source URL. It requires no authentication and cannot access organization files.

Query

curl https://www.siglata.com/ask \
  -H 'Content-Type: application/json' \
  -d '{"query":{"text":"How does the pilot work?","max_results":5},"prefer":{"mode":"list","response_format":"conv_search","streaming":false,"accept-language":"en-US"}}'

You can also use GET /ask?query=spreadsheets&streaming=false&lang=en-US. Portuguese is the default. For POST, set prefer.accept-language. The Accept-Language header applies when there is no explicit language preference.

Each result includes grounding.source_urls pointing to its source page. Ranking uses Orama text search over published content. The list mode retrieves pages; it does not generate language-model answers or execute actions.

Responses and limits

JSON: prefer.streaming: false or the Accept: application/json header.
SSE: prefer.streaming: true; the response sends one data: event containing results and closes the connection. SSE is the default without a JSON preference.
Queries are limited to 1,000 characters, request bodies to 8 KiB and results to 20; the default is 5 results.
Queries without matches return results: [].
Invalid request: 400; unsupported method: 405; oversized body: 413; wrong content type: 415; unavailable search: 503.

This implementation uses the list / conv_search profile of the NLWeb 0.54 models. Summaries, conversation history, persistent memory, cross-site search and other response formats are unsupported. POST fields outside the documented profile are rejected.

Discovery

The map follows the Schema Feeds 0.1 proposal, advertised in robots.txt. Each feed line contains one JSON-LD page with its language, text and source URL. Both formats are generated from the same public pages whenever the site is published.

The documentation also provides WebMCP for browser agents. The organization MCP at /v1/mcp is separate: it requires OAuth and preserves file-access scopes.