What it does

Deep Research is a Next.js web app that automates the grunt work of deep research: it searches the web, synthesizes sources, and spits out a structured report with citations. You bring your own LLM API keys and pick a search provider (Searxng, Tavily, Firecrawl, Brave, etc.). Everything runs in your browser by default, so your queries and drafts stay local unless you switch to server-side API mode.

The interesting bit

The project splits the workload between a “Thinking” model (planning and analysis) and a “Task” model (execution), which is a pragmatic way to balance quality against token costs. It also exposes the whole pipeline as an SSE stream or MCP server, so you can embed it into other tools rather than treating it as a siloed app.

Key highlights

• Supports a laundry list of LLMs: Gemini, OpenAI, Anthropic, DeepSeek, Grok, Mistral, Azure OpenAI, OpenRouter, Ollama, and generic OpenAI-compatible endpoints
• Can ingest local files (PDF, Office docs, text) to build a private knowledge base
• Offers WYSIWYM and Markdown editors for tweaking output, plus one-click knowledge graph generation
• Deploys to Vercel, Cloudflare Pages, Docker, or static hosting with minimal fuss
• PWA support means it behaves like a desktop app without an install step

Caveats

• Custom model lists only work in proxy mode, which is an odd restriction
• Docker images lag the latest release by 1–2 days, so you’ll see spurious update prompts
• The “about 2 minutes” claim for report generation is stated but not benchmarked in the README

Verdict

Worth a spin if you want a self-hosted research tool without vendor lock-in. Skip it if you need rigorous academic sourcing or don’t want to manage your own API keys and search provider accounts.