Open-source Deep Research
AutoSearch vs Perplexity API — Open-Source Deep Research
Choose the open-source path when you want to keep model choice, 40 channels, and Chinese sources.
01
Open-source vs closed
Perplexity API is a closed product with bundled answer generation. AutoSearch is MIT-licensed; ranking and citation logic are inspectable. You can audit how each result was retrieved, fork the channel set you need, and self-host without depending on a single vendor's roadmap.
02
LLM-decoupled vs bundled
Perplexity API ships its own model alongside retrieval. AutoSearch separates the two — your host (Claude Code, Cursor, AutoGen, custom) brings the LLM; AutoSearch returns cited evidence. Swap models without rebuilding your retrieval pipeline.
03
Multi-channel + Chinese sources
Perplexity covers general web search well; it does not cover Chinese sources natively. AutoSearch ships 40 channels including Zhihu, WeChat, Weibo, Xiaohongshu, Bilibili, and 36Kr — important for product research, policy tracking, and engineering decisions in Chinese markets.
How it fits
AutoSearch is the open-source alternative when Perplexity API's bundled-LLM approach blocks model swaps, when usage-based pricing makes large research runs expensive, or when you need Chinese source coverage. Hosts call AutoSearch through MCP; the LLM stays your choice; all citations carry channel + URL + date for downstream synthesis. Stick with Perplexity when you want a turnkey answer-generation API and don't need decoupled retrieval.
Try this prompt
Compare AutoSearch and Perplexity API for an agent that researches Chinese AI policy.
List what each tool can and cannot retrieve, with cited examples.