neuron-explorer
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
翻譯準備中
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證Official Querit integrations for AI agents: real-time, cited web search and page fetch for Pi, DSH, OpenCode, Claude Code, n8n, Zapier, and Browserbase.
翻譯準備中
取自儲存庫 README
正在載入完整 README。 以下為已儲存的摘錄。
Official Querit search integrations for automation and AI agent platforms.
Querit delivers real-time, authoritative web search results and clean page content for LLM applications. Each integration in this repository connects one platform to the Querit API (/v1/search + /v1/contents) for live web search and page fetching — with cited sources, domain/region/language/time-range filters, and output that is always treated as untrusted web data. The Oh My Pi entry below documents an upstream built-in provider, not a package in this repository.檢視完整 README ↗Official Querit search integrations for automation and AI agent platforms.
Querit delivers real-time, authoritative web search results and clean page content for LLM applications. Each integration in this repository connects one platform to the Querit API (/v1/search + /v1/contents) for live web search and page fetching — with cited sources, domain/region/language/time-range filters, and output that is always treated as untrusted web data. The Oh My Pi entry below documents an upstream built-in provider, not a package in this repository.
| Platform | Package | Install |
|---|---|---|
| Pi | pi-querit | pi install npm:pi-querit |
| DeepSeek Harness | dsh-querit | dsh plugin --profile web add dsh-querit |
| OpenCode | opencode-querit | "plugins": ["opencode-querit"] |
| Claude Code | claude-code-querit | /plugin marketplace add querit-ai/querit-plugins |
| n8n | n8n-nodes-querit | Install as an n8n community node |
| Oh My Pi (OMP) | Built-in Querit provider (not a plugin) | No install; available after upstream PR merges |
Two more integrations live in this repository but are not generally available yet:
| Platform | Directory | Status |
|---|---|---|
| Zapier | zapier-querit/ | Complete and locally validated; awaiting registration and Zapier public review |
| Browserbase | browserbase-querit-demo/ | Reference demo, not a published package |
How each package integration reads the API key and search defaults is covered in its own section below; the Oh My Pi section documents its upstream built-in provider. Recent updates are summarized in CHANGES.md.
Sign up on Querit.ai to get an API key with 1,000 free API calls per month — no credit card required. The same key works for all integrations.
A Pi extension that registers web_search and fetch_content as first-class tools. Your agent calls them automatically whenever an answer needs the live web.
Install
pi install npm:pi-querit
API key
Run the interactive setup wizard inside Pi once:
/querit-setup
The wizard validates the key, then lets you configure persistent search defaults (result count, time range, countries, languages, domain whitelist/blacklist) and an optional fixed-model auto-summary workflow. Everything except the API key can be skipped.
~/.pi/agent/querit-search.json (respects PI_CODING_AGENT_DIR), written with mode 0600 on POSIX.QUERIT_API_KEY environment variable. The environment variable takes precedence when both are present.See the pi-querit README for the full tool reference.
A provider package for the DeepSeek Harness ctx.web capability seam. It does not register the model-facing web_search tool; the agent preset does. By default, it reuses the official applyWebFetchTool helper to register web_fetch, and routes both tools through Querit-backed providers.
Install
dsh plugin --profile web add dsh-querit
Requires dsh 0.1.2+ — v1.1.2 keeps the 0.1.2 host seams (ctx.settings service, remote.credentials client namespace) and widens the peer range so 0.1.5-rc / 0.1.6-alpha hosts satisfy npm. On older dsh hosts pin [email protected].
Then wire it in the profile's cordis.patch.yml:
- insert:
- id: web-search-querit
name: 'dsh-querit'
config:
apiKeyEnv: QUERIT_API_KEY
- id: web
config:
searchProvider: web-search-querit
fetchProvider: web-search-querit
API key — resolved per operation in priority order:
QUERIT_API_KEY exported in the launching environment, or$DSH_HOME/.credentials.yaml (QUERIT_API_KEY: <key>, hot-reloaded), orapiKey on the web-search-querit row (least preferred — secrets should not live in composition files).Search defaults (count, timeRange, countries, languages, includeDomains, excludeDomains, includeContent, chunksPerDoc, fetch format/timeouts) are configured on the plugin row or in the hot-reloaded $DSH_HOME/settings.yaml web-search-querit: section.
See the dsh-querit README for the full config table.
An OpenCode v2 plugin (opencode ≥ 2.0.3) that registers Querit as a websearch provider — powering the built-in websearch tool, no OPENCODE_ENABLE_EXA/OPENCODE_ENABLE_PARALLEL flags needed — plus a web_fetch custom tool backed by Querit's crawler. On opencode v1 pin [email protected].
Install
Add the plugin to opencode.json:
{
"$schema": "https://opencode.ai/config.json",
"plugins": ["opencode-querit"]
}
Then restart OpenCode. On the first websearch call pick Querit in OpenCode's provider form, or pin it once with "websearch": { "provider": "querit" }.
API key — resolved per call in priority order:
QUERIT_API_KEY environment variable (default; rename via apiKeyEnv), orapiKey in the plugin options (least preferred — secrets should not live in config files):{
"$schema": "https://opencode.ai/config.json",
"plugins": [
{ "package": "opencode-querit", "options": { "count": 8, "timeRange": "m3", "languages": ["english"] } }
]
}
Search defaults and fetch behavior are configured the same way (count, timeRange, countries, languages, includeDomains, excludeDomains, includeContent, chunksPerDoc, fetchFormat, fetchCrawlTimeout, fetchMaxChars, maxOutputChars).
See the opencode-querit README for the full config table and local-testing instructions.
A Claude Code plugin that bundles a stdio MCP server exposing web_search and fetch_content, plus a /querit-ai:research skill for source-grounded answers with citations.
Install
/plugin marketplace add querit-ai/querit-plugins
/plugin install querit-ai@querit
API key — the plugin reads only the QUERIT_API_KEY environment variable, which the MCP server process inherits from the environment that launches Claude Code. There is deliberately no in-product key option: one variable covers local development, CI, and UI installs.
.mcp.json, source files, shell history, chat, or logs. Tool errors redact the active key before Claude sees them.Claude Code does not start a plugin's MCP server while a required option is unset, so configure api_key before expecting the Querit tools to appear in /mcp.
See the claude-code-querit README for the full tool reference and the safety limits.
An n8n community node with Search Web and Fetch Content operations, native paired-item and Continue On Fail handling, and AI-tool support.
Install n8n-nodes-querit through n8n's community-node interface, then create a Querit API credential and paste in the API key. n8n stores it as an encrypted credential; the node never writes it into workflow JSON or execution data. See the n8n-nodes-querit README for operation details and development instructions.
The pending upstream OMP change adds a built-in Querit provider for both its web_search provider chain and the URL-fetch/reader path used by the built-in read tool. This is not an npm plugin from this repository, so there is nothing to install here. Available after the upstream PR is merged.
Credentials
QUERIT_API_KEY in the environment that launches OMP, or/login querit in OMP and paste/save the key when prompted.Select Querit
After the upstream PR is merged, make Querit the first web_search provider and select it for URL fetching with:
omp config set providers.webSearchOrder '["querit"]'
omp config set providers.fetch querit
providers.webSearchOrder controls web_search provider priority, while providers.fetch controls the reader backend used when OMP's read tool fetches an HTTP(S) URL. After the upstream change lands, its search provider will send requests to POST https://api.querit.ai/v1/search with only { query, count } (count defaults to 10 and is clamped to 1..20), and its URL-content reader will send requests to POST https://api.querit.ai/v1/contents; both use Bearer authentication. Querit-specific language, site/domain, and date options remain unset so the service uses its defaults.
A Zapier Platform CLI integration with one Search action, Find Web Search Results. Authentication is a masked custom API-key field, tested with a fixed one-result POST /v1/search.
The integration is not published to Zapier's app directory yet. Registration, the private test push, and public review all require the company-owned Zapier developer account; the zapier-querit README carries the full handoff checklist.
A reference demo, not an installable package: it searches with Querit, then opens the top result in a Browserbase cloud browser through Playwright connectOverCDP and saves a screenshot as evidence. It needs both QUERIT_API_KEY and BROWSERBASE_API_KEY in the environment and deliberately does not read a .env file. See the browserbase-querit-demo README.
Each directory is an independent npm package with its own version, CI, and
release cycle. There is no root workspace: the integrations have independent
dependency trees, so each package keeps its own package.json,
lockfile, and node_modules.
querit-plugins/
├── .claude-plugin/ # Claude Code marketplace catalog
├── pi-querit/ # Pi extension (TypeScript, no build step)
├── dsh-querit/ # DeepSeek Harness provider package (compiled lib/ committed)
├── opencode-querit/ # OpenCode plugin (compiled lib/ committed)
├── claude-code-querit/ # Claude Code plugin (installable payload in plugin/)
├── n8n-nodes-querit/ # n8n community node (compiled for publishing)
├── zapier-querit/ # Zapier CLI integration (not published)
└── browserbase-querit-demo/ # Browserbase reference demo (not published)
Work inside a package directory; the root has no scripts. After npm ci, run that package's gates:
| Package | Gates |
|---|---|
pi-querit/ | npm run check && npm test |
dsh-querit/ | npm run check && npm run build && npm test |
opencode-querit/ | npm run check && npm run build && npm test |
claude-code-querit/ | npm run verify |
n8n-nodes-querit/ | install with npm ci --ignore-scripts, then npm run check && npm run lint && npm test && npm run build |
zapier-querit/ | npm run check && npm run build |
browserbase-querit-demo/ | npm run check && npm test && npm run build |
Every npm package runs its gates automatically on publish (prepack), so manual runs matter only while developing. The Claude Code marketplace catalog is validated separately from the repository root:
claude plugin validate . --strict
claude plugin validate ./claude-code-querit/plugin --strict
Publishing is per package. dsh-querit, opencode-querit, and pi-querit
publish automatically: run npm version <version> --no-git-tag-version in the
package directory (so package.json and package-lock.json stay in sync),
commit, and push to main. The per-package publish workflow skips pushes
whose version is already on npm and otherwise runs the package gates and
publishes with npm provenance. n8n-nodes-querit keeps its tag-triggered
flow: push an n8n-nodes-querit@<version> tag. All these workflows publish
through npm trusted publishing (OIDC, no NPM_TOKEN); each package needs a
one-time GitHub Actions trusted-publisher entry in its npm settings pointing
at its workflow file.
MIT — see each package's LICENSE file.