deep-research

English | 中文

A minimal “deep research” pipeline built with TypeScript + Bun + OpenAI API + SerpAPI.

Quickstart

1. Create .env from .env.example and fill keys.

2. Install deps:

bun install

3. Run:

bun run research "Explain the current state of solid-state batteries and commercialization challenges"

Output lands in out/report.md. It also writes out/report.state.json with the full intermediate state (plan/searches/sources/evidence/logs).

What it does

• Plans sub-questions + queries (OpenAI)
• Searches via SerpAPI
• Fetches pages and extracts text
• Extracts evidence items per source (OpenAI)
• Synthesizes a cited Markdown report (OpenAI)

CLI options

bun run research "..." --out out/report.md --max-urls 12 --timeout-ms 120000 --engine google --model gpt-5-mini

Increase timeout (useful for slower models):

bun run research "..." --timeout-ms 180000

Disable caching:

bun run research "..." --no-cache

Language

• Looking for Chinese docs? See README.zh-CN.md.