Advanced Nostr Search

A tool to help you search your favorite Nostr accounts.

Installation

npm ci
npm start

Open the local URL Vite prints (usually http://localhost:5173).

Note previews

Search results show quoted notes with their author, timestamp, text, and media. Quoted references and author names work with the existing Open In dialog. Web links show a title, description, and thumbnail when metadata is available. Unavailable notes and pages fall back to ordinary links. Each quote shows up to one web link preview with a smaller image; additional links and nested note references stay inline.

/api/preview runs with npm run dev and deploys as a Vercel function. No API key is required. The endpoint checks public destinations and redirects, pins DNS addresses, and limits fetch time and response size.

Ask AI

Each note has an Ask AI button. It opens an explanation of the note and lets readers ask follow-up questions, stop a response, or clear the conversation. Conversations stay in memory for the current page session. Include note images is on by default; turn it off for text-only models. Image URLs stay in the note context when images are off. Both this checkbox and Search the web when helpful are saved in the browser. Markdown links open in a new tab.

Generated answers default to the browser's preferred language (English when unavailable), regardless of the note's language. An explicit language request takes priority, followed by the language of the reader's follow-up question. This applies to both inference paths; panel labels stay in English.

Ask AI uses an installed Inference Bridge or another Inference Provider API extension first. No server credentials are needed for that path.

The optional hosted fallback uses the authenticated shared inference service through /api/inference, with Replit selected by default. Copy .env.example to .env.local for local use. Set these server-only values in Vercel Preview and Production:

INFERENCE_FALLBACK_ENABLED=true
SHARED_INFERENCE_PROVIDER=replit
REPLIT_INFERENCE_MODEL=google/gemma-4-26b-a4b-it
SHARED_INFERENCE_URL=https://ipa-inference-fallback-api.vercel.app/api/inference
SHARED_INFERENCE_APP_SECRET=<this app's existing secret>
SHARED_INFERENCE_REPLIT_API_KEY=<existing gateway key>

Keep selection/model/URL/enablement settings as Config and app/provider credentials as Secret. Never prefix them with VITE_ or commit secret values. The backend fixes the advancednostrsearch app identity and injects its selected provider key and model. Browser-selected providers, models, and credentials cannot override them. The proxy bounds requests and selects public fields; the shared service validates messages, images, search values, and options. Responses are non-streaming JSON; the proxy times out after 180 seconds within a 200-second Fluid compute function.

INFERENCE_FALLBACK_ENABLED=false disables hosted inference. Missing shared configuration fails closed. All hosted requests use the shared service; shared errors never trigger an automatic retry or a direct provider request. Shared Ollama and Gemini remain supported with their matching SHARED_INFERENCE_OLLAMA_API_KEY or SHARED_INFERENCE_GEMINI_API_KEY and backend-owned models (OLLAMA_MODEL and GEMINI_MODEL optionally override their defaults). Gemini search requires separate model validation and SHARED_INFERENCE_GEMINI_SEARCH_ENABLED=true. Legacy direct-provider keys, generic selection/model aliases, and quota settings are no longer read.

npm run dev serves both /api/inference and /api/preview. The shared service can run independently on port 3001: use SHARED_INFERENCE_URL=http://127.0.0.1:3001/api/inference in .env.local. Local HTTP is rejected in deployed Preview/Production.

Without an inference extension, the panel asks for consent before sending the note and questions to the hosted provider. Hosted requests retain a per-browser soft daily allowance of 50 successful explanations. Reasoning uses the provider default unless a caller supplies an explicit option. The shared service applies its own admission limits; browser counts can be reset and do not enforce an account-wide budget. If neither inference path is available, the panel links to Inference Bridge.

Checks

npm test
npm run build

For functional checks against a protected Preview deployment:

npm run smoke:inference -- --url https://PREVIEW_HOST/api/inference --provider replit --checks --followup --vercel
npm run smoke:inference -- --url https://PREVIEW_HOST/api/inference --provider replit --image-url 'https://images.unsplash.com/photo-1500534623283-312aade485b7?w=4000&q=100' --search --vercel

Each command makes a real generation; --followup adds another. --vercel uses the authenticated Vercel CLI for protected Preview access; omit it for public endpoints. VERCEL_CLI_PATH can point to the CLI's dist/vc.js. The image command exercises a full-size photo downloaded by the shared service, including its base64 upstream payload. Check the returned scene description; citations alone do not independently verify search execution.

The shared-Replit deployment verified on November 4, 2026 is dpl_2449EgdKfVVAu3Ayif2DsHKtcfZr (deployment URL). Restore that existing deployment through Vercel deployment history when rollback is needed, preserving its captured shared configuration and credentials.