In environments where noisy, platform‑specific conversations shape public perception, turning raw social posts into actionable insight requires more than single-model summaries. BettaFish treats analysis as a collective intelligence problem: lightweight agents specialize in search, multimodal parsing, database mining and report synthesis, then debate and refine findings in a forum-style loop to produce structured, reproducible reports.
What Sets It Apart
- Forum-driven multi-agent workflow — agents (Query, Media, Insight, Report) run in parallel and use a "forum"/moderator mechanism to iteratively debate, refine, and converge on explanations; this reduces single-model hallucination and yields more inspectable reasoning traces.
- End-to-end pipeline for Chinese social ecosystems — includes crawlers and parsers targeting Weibo, Xiaohongshu, Douyin, Kuaishou and comment pools, plus multimodal extraction for short-video platforms, which speeds domain-specific analysis out of the box.
- Modular, pure‑Python design with OpenAI‑compatible LLM adapters — easy to run in Docker, swap LLM endpoints, or integrate private data (via Insight Agent) for mixed public+private analysis scenarios.
- Built-in report IR → HTML/PDF rendering and lightweight sentiment/model toolkits — produces researcher-style final reports and sample analyses (example: Wuhan University report), handy for stakeholder briefings.
Who it's for & trade-offs
Great fit if you need reproducible, explainable social-media analysis focused on Chinese platforms or want a modular agent-based research pipeline you can customize. It’s also useful for research teams that want integrated crawling→analysis→reporting in one repository. Look elsewhere if you require enterprise-grade SLAs, commercial licensing (the repo is GPL‑2.0), or fully managed/scale-out crawling at web scale—some components (crawlers, models) require engineering to harden for heavy production traffic and legal compliance with target sites' terms.
Where It Fits
BettaFish sits between research prototypes (single-model analysis notebooks) and full commercial social-listening platforms: it lowers the integration effort for teams that can self-host and configure LLM APIs and databases, while offering deeper auditability than opaque SaaS analytics.
