Capabilities

GET /v1/capabilities returns a JSON snapshot of what the running instance supports. Clients and agents can call it on startup to adapt their behavior — instead of guessing what a self-hosted deployment has configured, ask.

Request

GET https://api.fastcrw.com/v1/capabilities
Authorization: Bearer YOUR_API_KEY

No request body. No query parameters.

Response Shape

{
  "version": "0.16.0",
  "llm": {
    "providers": ["anthropic", "openai", "deepseek", "openai-compatible", "azure"],
    "supportsBaseUrl": true,
    "serverKeyConfigured": false,
    "maxConcurrency": 0
  },
  "formats": {
    "supported": [
      "markdown", "html", "rawHtml", "plainText",
      "links", "json", "summary", "changeTracking"
    ],
    "changeTrackingModes": ["gitDiff", "json"]
  },
  "search": {
    "answer": true,
    "summarizeResults": true
  },
  "documents": {
    "parsers": ["pdf"],
    "fileUpload": {
      "supported": true,
      "endpoint": "/v2/parse",
      "maxBytes": 52428800,
      "types": ["application/pdf"],
      "ocr": false
    }
  }
}

All keys use camelCase.

Fields

version

string — The server binary's semver version (CARGO_PKG_VERSION).

llm

Field Type Description
providers string[] LLM provider tags the server's dispatch layer supports. Always ["anthropic", "openai", "deepseek", "openai-compatible", "azure"] — these are compile-time constants.
supportsBaseUrl boolean Always true. Any provider can be redirected to a custom base URL per request.
serverKeyConfigured boolean true when a server-side [extraction.llm] API key is configured. false means the instance has no server LLM key and requires callers to supply llmApiKey in the request body. Hosted SaaS sets this to false and relies on per-request BYOK.
maxConcurrency number Server-side fan-out cap for LLM calls. 0 when no server LLM config is present.
requireByokHeader string? Present only when the server requires a specific request header on LLM-touching calls. Absent when no header guard is configured.

formats

Field Type Description
supported string[] Output formats this instance can produce. The hosted API always returns all 8: markdown, html, rawHtml, plainText, links, json, summary, changeTracking.
changeTrackingModes string[] Diff modes available when changeTracking is in supported. Values: gitDiff, json. Empty on instances where changeTracking is unavailable.
Field Type Description
answer boolean Whether the instance supports LLM-synthesized answers on POST /v1/search.
summarizeResults boolean Whether per-result summaries are available on the search endpoint.
The `search` object does not expose whether a SearXNG backend is reachable. A live search configuration check happens at request time — if SearXNG is missing, the `/v1/search` endpoint returns HTTP 503 with `error_code: "search_disabled"`.

documents

Field Type Description
parsers string[] Active document parser types. ["pdf"] when the PDF feature is compiled in and [document] enabled = true in config. [] when PDF is disabled or not compiled.
fileUpload.supported boolean true when POST /v2/parse accepts uploads. Matches parsers non-empty.
fileUpload.endpoint string Always /v2/parse when uploads are supported.
fileUpload.maxBytes number Maximum accepted upload size in bytes (default 52 428 800 = 50 MiB). Set by [document] max_upload_bytes in config.
fileUpload.types string[] Accepted MIME types. ["application/pdf"] when PDF is enabled, [] otherwise.
fileUpload.ocr boolean Always false. The built-in PDF parser does not include an OCR engine — scanned or image-only PDFs yield empty or partial text.

Hosted API Values

On api.fastcrw.com you can expect:

  • llm.serverKeyConfigured — false (SaaS relies on per-request BYOK)
  • documents.parsers — ["pdf"] (PDF support is enabled)
  • documents.fileUpload.maxBytes — 52428800 (50 MiB)
  • formats.supported — all 8 formats
  • search.answer and search.summarizeResults — true

Using Capabilities at Runtime

Agents and SDKs

Call GET /v1/capabilities once on startup and cache the result. Use it to:

  • decide whether to send formats: ["json"] (check formats.supported contains "json")
  • decide whether to supply llmApiKey in the body (check llm.serverKeyConfigured)
  • decide whether to offer file-upload UI or API paths (check documents.fileUpload.supported)
  • confirm whether changeTracking is available before using monitor scrapes (check formats.supported contains "changeTracking")
import httpx

caps = httpx.get(
    "https://api.fastcrw.com/v1/capabilities",
    headers={"Authorization": "Bearer YOUR_API_KEY"},
).json()

if caps["llm"]["serverKeyConfigured"]:
    llm_key_in_body = None          # server has a key
else:
    llm_key_in_body = "sk-..."      # caller must supply BYOK

pdf_ok = caps["documents"]["fileUpload"]["supported"]

Self-Hosted Instances

Self-hosted deployments differ from the hosted API in several ways a caller cannot predict without checking:

  • PDF parsing may be disabled ([document] enabled = false) or not compiled in
  • A server-side LLM key may or may not be configured
  • The [document] max_upload_bytes limit can be set below 50 MiB
  • SearXNG may not be connected

Always call /v1/capabilities before making assumptions about what a self-hosted instance can do.

Authentication

GET /v1/capabilities is an authenticated route on both hosted and self-hosted deployments (when API keys are configured). The response contains no secrets, but it describes deployment configuration, which is worth protecting.

On a self-hosted instance with no API keys configured, the route is public.

  • Output Formats — the 8 formats and when to use each
  • PDF Parsing — the POST /v2/parse upload endpoint
  • Search — POST /v1/search and the LLM answer path
  • Self-Hosting — configuration reference for [document], [extraction.llm], and [search]