Skip to content
Conduit

skill

tanso

Agent-friendly AI Search CLI for Chinese web signals, powered by Bocha, Zhihu, and Volcengine Ark.

geekjourneyx/tansocommunitylistedunreviewednot running here yet

Set up

Skills are installed by placing their folder where your agent loads skills.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below; one-click install arrives with accounts.

README

synced Oct 14, 2026

Tanso

Search Chinese web sources from one Go CLI.

License: MIT Go

Tanso queries Bocha, Volcengine Ark, and Zhihu through provider APIs, then returns normalized terminal output or automation-safe JSON. The name comes from the feel of "探索 / 探索する": Tanso explores Chinese internet signals for developers, AI agents, and research workflows without scraping, browser sessions, or hidden side effects.

Install

Recommended:

npm install -g @geekjourneyx/tanso
tanso version

The npm package installs the matching GitHub Release binary for your platform and verifies it against SHA256SUMS.

Alternative Go install:

go install github.com/geekjourneyx/tanso/cmd/tanso@v2.0.2

Prebuilt binaries and checksums are available on the GitHub Releases page.

From a local checkout:

make build
./tanso version

Configure

Initialize the default config file:

tanso config init
tanso config path

This creates a config file at Tanso's stable user config path:

PlatformDefault path
macOS/Linux~/.config/tanso/config.yaml
Windows%USERPROFILE%\.config\tanso\config.yaml

If XDG_CONFIG_HOME is set, Tanso uses $XDG_CONFIG_HOME/tanso/config.yaml.

Run tanso config path for the exact path on your machine.

Edit that file and set the credentials for the providers you want to use:

ProviderEnvironment variables
BochaBOCHA_API_KEY
Volcengine ArkARK_API_KEY or VOLCENGINE_API_KEY
ZhihuZHIHU_ACCESS_SECRET or ZHIHU_API_KEY

The config file uses the same provider fields:

bocha:
  api_key: ""

volcengine:
  api_key: ""
  model: doubao-seed-2-0-lite-260215

zhihu:
  access_secret: ""

Configuration path resolution is:

  1. --config
  2. TANSO_CONFIG
  3. $XDG_CONFIG_HOME/tanso/config.yaml, when XDG_CONFIG_HOME is set
  4. ~/.config/tanso/config.yaml

Configuration value precedence is:

  1. Provider environment variables
  2. Config file values
  3. Built-in defaults

Inspect the merged config without leaking secrets:

tanso config show --json

tanso sources --json is an inventory endpoint. Its configured field only means local credential material is present after config/env resolution; it is not a live authentication check. Use a real source query to verify whether credentials work.

Before publishing a release, verify version alignment across package.json, Makefile, CHANGELOG.md, and the release tag:

make version-check
make release-check

Quick Start

tanso version
tanso --help
tanso sources --json
tanso skills list --json
tanso skills read tanso --json

tanso "AI Agent 商业化" --json
tanso all "AI Agent 商业化" --json
tanso "AI Agent 商业化" --source bocha_web --source zhihu_search --json
BOCHA_API_KEY=... tanso bocha "AI Agent 商业化" --json
BOCHA_API_KEY=... tanso bocha "微信 AI Agent" --json
ARK_API_KEY=... tanso volc "瑞幸咖啡 2026 门店数是否可信" --json
ZHIHU_ACCESS_SECRET=... tanso zhihu "AI 搜索" --json
ZHIHU_ACCESS_SECRET=... tanso zhihu web "ChatGPT 桌面版" --json
ZHIHU_ACCESS_SECRET=... tanso zhihu hot --json

Human output defaults to a table. Use --json for scripts, agents, CI, and downstream tools. Use tanso --help or tanso <command> --help to inspect the installed CLI surface.

Agent Skill

Recommended Agent Skills install:

npx skills add geekjourneyx/tanso

Tanso also ships the same Agent SOP with the CLI package so agents can read instructions that match the executable on PATH:

tanso skills list --json
tanso skills read tanso
tanso skills read tanso --json

skills read defaults to raw Markdown for direct agent context. With --json, the same content is returned in the content field with version metadata. Use this when you need to verify the SOP bundled with the installed CLI.

Output

Retrieval commands return a stable envelope. A successful JSON response looks like this:

{
  "version": "2.0.2",
  "query": {
    "text": "AI Agent 商业化",
    "mode": "search",
    "sources": ["bocha_web"],
    "limit": 10
  },
  "status": "ok",
  "results": [
    {
      "source": "bocha_web",
      "title": "Example result title",
      "url": "https://example.com/article",
      "snippet": "A normalized summary from the provider response."
    }
  ],
  "source_status": [
    {
      "source": "bocha_web",
      "status": "ok",
      "results": 1,
      "effective_limit": 10,
      "duration_ms": 842,
      "error": null
    }
  ],
  "errors": []
}

Exit codes are part of the public contract:

CodeMeaning
0Success
1Partial success
2Invalid argument
3Config error
4Credential missing
5Source error
6Timeout
7No results
9Internal error

Sources

Source IDCommandProviderCapability
bocha_webtanso bochaBochaWeb search
volcengine_answertanso volcVolcengine ArkWeb-grounded answer
zhihu_searchtanso zhihuZhihuIn-site search
zhihu_webtanso zhihu webZhihuGlobal web search
zhihu_hottanso zhihu hotZhihuHotlist

Zhihu global search supports provider-specific filters:

tanso zhihu web "ChatGPT 桌面版" \
  --filter 'host=="example.com"' \
  --search-db realtime \
  --json

--filter and --search-db are only valid for tanso zhihu web.

Automation Contract

Tanso keeps the automation contract narrow and predictable:

  • stdout is reserved for results.
  • stderr is reserved for diagnostics.
  • JSON output keeps source IDs, source status, error codes, and exit codes stable.
  • Provider behavior stays behind typed Go adapters.
  • Built-in timeout defaults to 45s; override it with --timeout when a workflow needs a different budget.

Development

make build
make test
make lint
make release-check

The normal test suite does not require real provider credentials. Real API smoke checks are separate:

make smoke-bocha
make smoke-volcengine
make smoke-zhihu

Non-Goals

Tanso v1.0.0 intentionally does not implement browser scraping, cache, reranking, plugin runtime, MCP, stdin query input, Bocha image search, or Zhihu direct answer. Those boundaries keep the CLI small, testable, and stable.

License

MIT

Author

↑↓ navigate openesc close