Skip to content
Projects

shenyangs/Guanlan has 57 GitHub stars and 3 forks. GitHot measured 0 in 7 days and 0 in 15 days. Last checked 2026-11-18 10:34 UTC. Last push 2026-11-03.

How these counts are measured

Measured star growth

in 24h
Update delayed
in 7 days
0
in 15 days
0
in 1 month
Collecting history

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

观澜 / Guanlan:AI Agent 的中文互联网研究、阅读与信源路由工具。

AI

docs/README_en.md

Original

Guanlan

Observe the current, trace the source, keep the boundary clear.

Guanlan is a CLI-first Chinese-web research and source-routing tool for AI agents. It is not just a search wrapper: it routes a question to suitable source pools, reads public pages, tracks hot topics, builds evidence packets, and tells downstream agents what can be treated as official evidence, media reporting, user samples, or only open-web context.

The canonical documentation is maintained in Chinese. This English page is a synchronized summary for international users and agent integrators.

Start Here

Document Use it for
Chinese README Canonical positioning, install path, examples, and release notes.
Changelog Version-by-version changes and boundary adjustments.
Agent Playbook Durable agent memory: dynamic workflows, benchmark discipline, fallback rules.
Agent usage guide Search/read/hotnews/archive/MCP usage for AI agents.
Contract Stable output fields for CLI/MCP/HTTP/RAG integrations.
Local LLM guide Let Ollama, LM Studio, Open WebUI, and local models use Guanlan as a web evidence provider.
Troubleshooting Keychain prompts, network failures, cookies, and platform fallbacks.
Source attribution Open-source projects and public sources referenced by Guanlan.

What Is Stable Now

  • Public web search with China-aware profiles, source classification, evidence roles, source cards, quality scoring, and default broad result pools.
  • URL reading with Jina Reader first, direct HTML fallback, quality reports, strict mode, batch reading, caching, and local snapshots.
  • Chinese hot lists and trend clustering via hotnews, including multi-source status and stale/cache signals.
  • Structured research workflows: research, compare, timeline, and dossier turn search results into evidence packets, comparisons, event timelines, and entity dossiers.
  • Local archive and RAG bridge: SQLite/FTS archive, ingest audit, prompt-ready context, wiki export, and LangChain/LlamaIndex/OpenWebUI packs.
  • Agent integration: CLI-first usage, optional MCP, optional local read-only HTTP service, and prompt/context commands for local models.

v0.4.1 Search Reliability Updates

  • Query guard: meaningless or keyboard-mash queries are rejected before hitting search backends, with diagnostics instead of random pages.
  • Query rewrite: short factual queries, short reputation/ecommerce queries, and overlong queries are conservatively expanded or compressed.
  • Recovery search: DuckDuckGo fallback passes and multi-entity fan-out reduce empty or one-entity-only result sets.
  • Agent workflow plan: quality summaries now explain whether a task should follow a 2-step, 3-step, or 4-step Guanlan workflow before falling back to generic web search.
  • Empty JSON results can include structured diagnostics, so agents can report “query needs rewriting” instead of “search failed”.

Core Commands

guanlan capabilities
guanlan welcome
guanlan search "query" --profile china --limit 80
guanlan search "policy topic" --profile china --scope party_central --trace
guanlan search "product reviews" --profile china --scope ecommerce --format context
guanlan route "mixed Chinese research task" --json
guanlan research "query" --profile china --advisor
guanlan compare "A" "B" --focus "price reputation risk" --limit 80 --format context
guanlan timeline "event latest progress" --limit 80 --format context
guanlan dossier "company or product" --focus "business reputation risk" --limit 80 --format context
guanlan read "https://example.com/article" --quality-report
guanlan hotnews today --limit 80 --trends
guanlan feeds curated --limit 80
guanlan archive ingest-research "query" --limit 80
guanlan archive context "query" --limit 20
guanlan doctor --trace

Agent Workflow Rule

Do not reduce Guanlan to a single generic search call.

  • 2-step: search -> read when results are already usable and you only need to verify a representative original page.
  • 3-step: route -> research -> scoped search for normal research tasks or when quality signals are not yet strong enough.
  • 4-step: add hotnews for recent/hot tasks, feeds for tech/AI/developer tasks, or compare/timeline/dossier when source diversity is too narrow.

Only fall back to generic web_search / web_fetch after the appropriate Guanlan workflow has been completed and still lacks key evidence.

Safety Notes

Guanlan is not a stealth crawler and not an account automation framework. It is read-first, low-disturbance, source-aware, and explicit about authentication. Public sources are preferred. Cookie, browser, Keychain, and login-state access must be requested by the user. Guanlan does not post, comment, like, follow, or send messages automatically.