Search and Ask AI

svedocs creates search records for each page and section. Local MiniSearch, Algolia, Typesense, Cloudflare indexing, and Ask AI citations can all work from those records.

Runtime routes

TypeScriptsrc/routes/api/search/+server.ts
1import { createConfiguredSearchResponse } from 'svedocs/search';2import config from 'virtual:svedocs/config';3import records from 'virtual:svedocs/search';45export const GET = ({ platform, request }) => {6  return createConfiguredSearchResponse(config, records, request, {7    env: platform?.env ?? process.env8  });9};

Generated templates use this runtime route. It reads svedocs.config.ts and selects local, algolia, typesense, or cloudflare-ai-search. If credentials or bindings are missing, the route uses local MiniSearch instead.

TypeScriptsrc/routes/api/ask/+server.ts
1import { createConfiguredAskResponse, createMemoryRateLimiter } from 'svedocs/ai';2import config from 'virtual:svedocs/config';3import records from 'virtual:svedocs/search';45const rateLimiter = createMemoryRateLimiter({ windowMs: 60_000, max: 30 });67export const POST = ({ platform, request }) => {8  return createConfiguredAskResponse(config, records, request, {9    env: platform?.env ?? process.env,10    rateLimiter11  });12};

createConfiguredAskResponse selects mock, cloudflare-ai-search, cloudflare-workers-ai, or openai-compatible from the config. During development, missing credentials produce a mock answer with local citations instead of an error.

Local search

Local search requires no external service. MiniSearch ranks matches from titles, sections, paths, body content, and metadata. Use the locale and kind query parameters to limit the results.

Text
1/api/search?q=cloudflare&locale=zh

Algolia

Set search.provider = 'algolia' and provide ALGOLIA_APP_ID, ALGOLIA_SEARCH_KEY, and ALGOLIA_INDEX_NAME in the server runtime environment. Algolia is optional and server-routed, so search keys do not need to be embedded in the default theme. Use it when an existing DocSearch crawler or hosted search pipeline already owns your index.

Typesense

Set search.provider = 'typesense' and provide TYPESENSE_HOST, TYPESENSE_SEARCH_KEY, and optionally TYPESENSE_COLLECTION. Typesense suits teams that want a self-hosted search service with fast typo tolerance. svedocs calls it over REST, so the generated site does not gain another runtime package.

Cloudflare AI Search

This integration supports the current ai_search and ai_search_namespaces bindings. It is opt-in; new sites use local MiniSearch.

Indexing

Shell
1svedocs index --provider cloudflare-ai-search --dry-run2svedocs index --provider cloudflare-ai-search --strategy replace --wait

The index command uploads Markdown documents, can delete stale records explicitly, and reports failures per record. Display data is stored in a compact svedocs JSON field, while locale and kind remain available for filtering.

Ask AI

The Ask AI route supports JSON and event-stream responses. Cloudflare AI Search can pass through its token stream when the binding supports chatCompletions({ stream: true }); other providers return structured answer and citation events. Requests may include locale or kind, and the default theme sends the active filters automatically.

OpenAI-compatible Ask AI

Set ai.provider = 'openai-compatible' and provide OPENAI_COMPATIBLE_API_KEY, OPENAI_COMPATIBLE_MODEL, and optionally OPENAI_COMPATIBLE_BASE_URL. svedocs builds a short RAG prompt from local search records and sends it to any compatible Chat Completions service, including OpenAI, OpenRouter, Groq, Together, or a private gateway.

When you need lower-level control, import createAlgoliaSearchProvider, createTypesenseSearchProvider, createCloudflareAiSearchProvider, createWorkersAiProvider, or createOpenAiCompatibleProvider directly and pass the resulting provider to createSearchResponse or createAskResponse.

Rate limits

TypeScript
1import { createCloudflareKvRateLimiter } from 'svedocs/ai';23const rateLimiter = createCloudflareKvRateLimiter({4  namespace: platform.env.SVEDOCS_RATE_LIMIT,5  windowMs: 60_000,6  max: 307});

Use the memory limiter for local development and KV-backed limiting for edge deployments.