argoB61

Multilingual search infrastructure built for AI agents: language detection → domain routing → multi-engine recall → RRF fusion → evidence appraisal, delivering compact JSON evidence candidates with a credibility breakdown instead of a human-facing summary page or link list. One command on DSH gives the model ten `mcp__argo__*` tools.

Star

81+12

7-day growth

Contributors

Language

Python

Updated

2026-08-13

Not an installable plugin package

The package.json declares no dsh.bundle, so even once installed into a profile it is never assembled. It may be an index list, documentation, a standalone app, or still in development — check the repo README for how to use it.

Editor's notes

Model-native search, AI search and metasearch all solve "a person looking for information"; Argo solves "an agent looking for information" — and the difference is not the interface, it's the deliverable. People get a summary page or a SERP link list. An agent should get evidence it can rank, re-check with `fetch`, and consume without blowing up its context. Argo treats search as an evidence pipeline: detect the language, route by domain to the right sources, recall across multiple engines, fuse with RRF, then appraise the evidence and emit compact JSON — evidence candidates plus a credibility breakdown across selection, absorption, freshness and consensus. Vertical questions (market data, chemical formulas and the like) go straight to vertical sources for a direct answer rather than scanning web page titles. Repeat queries hit a two-layer cache (in-memory plus SQLite), putting hot queries at roughly 10ms. Cost is handled by a budget mode that prefers free sources, with every API key optional. Coverage spans Chinese, English, academic, code, shopping, finance, news and encyclopedic sources, and web search works alongside local file search. DSH users can install the `.dsh-plugin` bundle directly — one command and the model has ten `mcp__argo__*` tools.

▸An evidence pipeline — language detection → domain routing → multi-engine recall → RRF fusion → appraisal — emitting compact JSON, not a link list
▸Credibility is broken out across selection, absorption, freshness and consensus, so the agent can rank and re-verify with `fetch`
▸Vertical questions route straight to vertical sources for direct answers; a two-layer cache (memory + SQLite) keeps hot queries near 10ms
▸Budget mode prefers free sources and every key is optional; `dsh plugin --profile web add` installs ten `mcp__argo__*` tools in one line

Score breakdown

Scored 2026-08-14

Activity

17.6

Star heat

23.1

Engineering

Maintainer

5.3

五语言文档的 Agent 搜索工具，120+ 引擎+MCP；通用工具而非 DSH 专属插件

Share card

Drop it into your README — it links back to this page.

Badge (inline)

[![dshfind](https://dshfind.com/api/badge/taxueseek/argo)](https://dshfind.com/en/plugins/taxueseek/argo?ref=badge)

Card (440×200)

[![dshfind](https://dshfind.com/api/card/taxueseek/argo)](https://dshfind.com/en/plugins/taxueseek/argo?ref=badge)

#dsh-ecosystem