[heydryft]

What it adds

Outcrawl (https://outcrawl.ai) as a keyed web search engine, built the same way as Brave (#1650) and Firecrawl: an outcrawl search platform whose key lives in the normal key pool, plus an adapter in services/search.ts. Outcrawl is a search and scrape API for AI agents. Its free tier is 1,000 credits a month, no card needed, which fits the catalog's "free to start without a credit card" rule.

• server/src/providers/outcrawl.ts: OutcrawlProvider (search only). Validation reads the account's credit balance (GET https://api.outcrawl.ai/v1/credits with the bearer key), which spends nothing, like the Firecrawl and Linkup probes. It goes through the shared probe, so 401/403 is a bad key, other 4xx still proves the key, and 5xx or a timeout is inconclusive.
• outcrawl adapter in services/search.ts: sends POST /v1/search with { query, limit } and maps each hit's title, url and snippet to { title, url, content } through keepHits. Outcrawl returns hits already ranked and with no score, so keepHits scores them by position. It has no news vertical, so a news search runs as a web search (same as Keenable). Error handling needs nothing special: 401 benches the key, 402 (out of credits) takes the long credit cooldown, 429 the rate-limit cooldown, all through the existing coolDown.
• outcrawl appended to SEARCH_PLATFORMS (shared, server, client), the Platform union, the keys route, key-parser (OUTCRAWL_ prefix and auth.json), SEARCH_PLATFORM_INFO, the Add key picker ("Outcrawl (1,000 credits/month)", linking to https://outcrawl.ai/console/keys) and the platform color map. Every list keeps its order, with Outcrawl after Brave.

Defaults

Opt-in only. No existing engine, order or default changes. Like every engine, it only runs once the catalog has a searchEngines row for it, so the row, its priority and its Premium/Free timing are your call. A row like { platform: 'outcrawl', engineId: 'outcrawl-search', displayName: 'Outcrawl Search', quotaLabel: '1,000 credits/month', features: { news: false, answer: false, fullContent: false } } matches what the adapter does.

What is sent to Outcrawl

The search query, limit, and the API key in the Authorization header. Nothing else.

How I tested

• New tests, mirroring the Brave and Firecrawl ones:
  • providers/search-providers.test.ts: Outcrawl added to the shared validation cases (no-cost probe URL and method, 401 invalid, other 4xx authenticated, 5xx and timeout inconclusive, chat refused) and to the registry check.
  • services/search.test.ts: request URL, bearer header and body, hit mapping (entity decoding, hits without a URL dropped, title falling back to the URL, rank scores), a 401 benching the key, and a 402 parking it for the credit cooldown.
• npx vitest run on search.test.ts, search-providers.test.ts, catalog-sync.test.ts and keys.test.ts: 201 passed.
• Full server suite (npm run test -w server): 382 files passed, 4818 tests passed, 6 skipped. One timing test failed in that run (compression.test.ts, "stays within the 100 KB synchronous performance budget"); it passes when run on its own (21 of 21) and does not touch search.
• Client: npx vitest run 436 passed (56 files), tsc -b clean, check:i18n passed for 60 locales.
• npm run lint clean; tsc --noEmit (server) clean.
• Live, through the new code path: an in-memory DB with one Outcrawl key and an outcrawl-search engine row, then OutcrawlProvider.validateKey(key) returned true and runSearch('open source LLM API gateway', { engine: 'outcrawl-search', maxResults: 5 }) returned provider: outcrawl, 5 results from llmgateway.io (2), github.com (2) and litellm.ai, each with a title and snippet.

Disclosure: I work on Outcrawl.

This PR was prepared with AI assistance (omp with Claude): the code, tests and this description were written with it, and the tests and the live call above were run as part of preparing it.