Deep research reports from any LLM you please

What it does The app is essentially a pipeline: it queries Google or Bing, uses JinaAI to scrape the contents of the top results, then hands that text to an LLM of your choosing—Gemini, GPT-4, Claude, DeepSeek, or a local Ollama instance—to generate a formatted report. You can also upload local TXT, PDF, or DOCX files and skip the web entirely. Finished reports are stashed in a browser-side Knowledge Base for later retrieval.

The interesting bit The “Flow” feature treats research as a visual tree rather than a one-off query. It auto-generates follow-up questions from a report, lets you branch into recursive sub-topics, and eventually consolidate multiple related reports into a single summary. It is basically mind-mapping for people who prefer citations.

Key highlights

• Swappable AI backends: supports Google, OpenAI, Anthropic, DeepSeek, OpenRouter, and local Ollama models, configurable in lib/config.ts.
• Recursive research trees via the Flow interface, with drag-and-drop visual mapping and report consolidation.
• Local file ingestion: mix uploaded TXT/PDF/DOCX files with web sources, or run entirely offline with local documents and Ollama.
• Rate limiting is optional and Redis-backed; disabling it lets you run without extra infrastructure.
• Exports to PDF, Word, and plain text; a Knowledge Base persists reports in browser local storage.

Caveats

• The Knowledge Base lives in browser local storage, so your archive evaporates if you clear site data.
• Reasoning models like OpenAI o1 or DeepSeek R1 may hit serverless function timeout limits on default Vercel tiers; the README warns you to bump maxDuration.
• It is a TypeScript web app that requires bringing your own API keys for search and most LLM providers; there is no free bundled tier.

Verdict Worth a spin if you want structured research tooling but refuse to lock yourself into a single vendor’s black box. Skip it if you are looking for a fully managed, zero-configuration research assistant.