Chinese web intelligence for autonomous agents

观澜 Guanlan

Clean evidence packets for agents navigating the Chinese internet.

Guanlan research packet
Prompt

AI 出海政策与平台反馈

Agent asks one question. Guanlan decides which source surfaces should be trusted, sampled, or treated only as open-web context.

guanlan research "国产 AI 应用 出海" --profile china --limit 80 --advisor
Official policy
Media industry
Community feedback
Evidence role Source mix guard: official + media + user sample
ready
Freshness Recent signals separated from evergreen references
clear
Advisor Next query suggested before the agent overclaims
bounded
01 / Flow

One visual language, one research flow.

Route

Pick the waterway.

Scope and profile select source families before the agent starts reading.

Read

Clean the source.

Public pages become compact source-aware context with quality notes.

Research

Shape the packet.

Source mix, freshness, and advisor notes arrive in one usable output.

02 / Run

Designed for the agent’s next step.

Every response is shaped so another agent, MCP client, or local model can continue without guessing what the evidence means.

guanlan compare "产品A" "产品B" --focus "价格 口碑 风险" --limit 80 --format context
China-aware Profiles and scopes mirror Chinese web structure.
Agent-ready Markdown, JSON, MCP, and workflow outputs.
Traceable Quality signals explain when evidence is thin.
Read-only Public search and reading before authenticated actions.
03 / Install

Install the river into your agent.

Use one clean install path, then smoke test before wiring it into your agent.

uv
uv tool install --force --upgrade guanlan
guanlan version
guanlan doctor --install-check
Homebrew
brew update
brew reinstall shenyangs/tap/guanlan
guanlan hotnews today --limit 80