sunrisefromdark/agentradar

AgentRadar is an open-source tool for tracking trends in the AI agent software space. It monitors public signals from GitHub and other sources, scores and ranks projects, and produces structured daily and weekly reports so that researchers, developers, and product teams can follow what is genuinely gaining traction rather than just reacting to one-day spikes on trending lists. The system is organized as a pipeline of specialized agents, each with a clear job. A signal collection agent gathers raw data from public sources. A normalization and scoring agent standardizes data from different sources into a common format and produces a ranked list with explained scores. A daily report agent writes a digest of what appeared that day, including summaries, reasons for inclusion, and risk notes. A weekly trend agent looks across a seven-day window to distinguish lasting trends from short-term noise. An observer agent watches early-stage projects that have not yet reached mainstream visibility. A knowledge card agent condenses high-value projects into structured notes for later reference. The output files land in a data directory with predictable naming: daily JSON and markdown reports, weekly judgment files, knowledge cards, and observer pool entries. A local read-only web console lets you browse these outputs from a browser without needing to deploy anything. There is also a hosted version at app.agentradar.top for people who want to see results before deciding whether to run it locally. Installation requires Node.js with pnpm. After copying the environment file and running the install step, you can launch the local web console with one command. Running the full pipeline follows a straightforward sequence: a daily run, a verification step, then the weekly rollup. The README is in Chinese with a linked English version. The open-source repository contains the data pipeline and local console, login and personalization features are available only in the hosted version.