HN 표시: 오픈 소스 Perplexity 복제 파일 백엔드, 스트리밍 답변

Search the web, synthesize answers, cite sources. Like Perplexity — but open source, and the entire backend is one file. Built with Next.js and OnCell. - Web search — searches the internet for every query via Tavily - Synthesized answers — LLM reads the sources and writes a structured response - Source cards — every answer shows the web pages it came from, with favicons and links - Follow-up questions — conversation context is maintained per session - Past research — all queries are indexed and searchable for future reference - Zero infra — search history, conversation memory, and agent runtime are all handled by OnCell A research agent needs a runtime, a database for conversation history, and a search index for past research. That's three services. OnCell gives you all of that in one call: const cell = await oncell.cells.create({ customerId: "research", agent: agentCode, }); // cell.db → conversation history per session // cell.search → index and search past research // cell.fetch → web search via Tavily // One file. Zero config. git clone https://github.com/oncellai/oncell-research.git cd oncell-research npm install - OnCell — oncell.ai - OpenRouter — openrouter.ai - Tavily — tavily.com (free tier: 1000 searches/month) Add them to .env.local : cp .env.example .env.local # Edit with your keys node scripts/setup.js npm run dev Open http://localhost:3000 and start researching. Next.js ├── / Research UI (Perplexity-style) ├── /api/chat POST → OnCell cell OnCell cell ├── db Session history per user ├── search Index of past research └── agent Web search → LLM synthesis → cited answer The agent code is one file: lib/agent-raw.js . It searches the web via Tavily, builds a prompt with the results, calls the LLM, saves history, and indexes the research for future reference. Change the LLM — set LLM_MODEL in .env.local (default: google/gemini-2.5-flash ) Change search depth — edit search_depth and max_results in lib/agent-raw.js Style the UI — edit app/page.tsx . Standard Next.js + Tailwind. - OnCell — build AI agents without building infra - oncell-support-agent — another open source OnCell project Apache-2.0