
kdush/openclaw-mirosearch
Developer toolsこのページは AI により自動翻訳されています。技術的な詳細は原文をご確認ください。 原文を見る
Diting (谛听) is an open-source agentic research service built on MiroThinker. It combines multi-provider web search, full-page extraction, multi-step reasoning, source verification, asynchronous...
概要
Diting (谛听) is an open-source agentic research service built on MiroThinker. It combines multi-provider web search, full-page extraction, multi-step reasoning, source verification, asynchronous execution, and structured Markdown reports. Current stable release: . See the Changelog for released changes and the Roadmap for planned work. - Six research modes: balanced, verified, research, production-web, quota, and thinking. - Six search profiles: searxng-first, serp-first, multi-route, parallel, parallel-trusted, and searxng-only. - Search providers: SearXNG, SerpAPI, Serper, Tavily, and optional Sogou integration. - Full-page extraction for HTML, text, PDF, JSON, RSS, Atom, and XML, with redirect SSRF checks, response-size limits, encoding recovery, table preservation, and natural-boundary truncation. - FastAPI task service backed by arq and Valkey, with persistent task metadata, event streams, results, cancellation, and result caching.
README
以下は原文 README の抜粋です。原文を見る。
Diting (谛听)
Diting (谛听) is an open-source agentic research service built on MiroThinker. It combines multi-provider web search, full-page extraction, multi-step reasoning, source verification, asynchronous execution, and structured Markdown reports.
Current stable release: v0.2.11. See the Changelog for released changes and the Roadmap for planned work.
Current capabilities
- Six research modes:
balanced,verified,research,production-web,quota, andthinking. - Six search profiles:
searxng-first,serp-first,multi-route,parallel,parallel-trusted, andsearxng-only. - Search providers: SearXNG, SerpAPI, Serper, Tavily, and optional Sogou integration.
- Full-page extraction for HTML, text, PDF, JSON, RSS, Atom, and XML, with redirect SSRF checks, response-size limits, encoding recovery, table preservation, and natural-boundary truncation.
- FastAPI task service backed by arq and Valkey, with persistent task metadata, event streams, results, cancellation, and result caching.
- Gradio UI with API-backed reconnect, progress display, clickable citations, and Markdown/PDF/Word export.
- OpenAI-compatible and Anthropic model clients, stage-specific model routing, key rotation, bounded retries, and model failback.
- Fail-closed FastAPI authentication, request rate limiting, caller-scoped cancellation, and per-task tool isolation.
Planned features such as batch scrape_urls, Prometheus/Grafana, RRF ranking, an MCP adapter, and Helm packaging are not implemented yet; they are tracked in the Roadmap.
Architecture
The default Docker Compose stack contains five services:
| Service | Purpose | Default host port |
|---|---|---|
app |
Gradio UI and compatibility API | 8080 |
api |
FastAPI research API | 8090 |
worker |
arq research worker | — |
valkey |
Queue, task state, event streams, and cache | internal only |
searxng |
Self-hosted search provider | 27080 |
See Architecture for data flow and module boundaries.
Quick start
Requirements: Docker with Compose v2, an OpenAI-compatible or Anthropic LLM endpoint, and at least one usable search provider.
cp .env.compose.example .env.compose
# Edit .env.compose and replace placeholder LLM/search credentials.
docker compose --env-file .env.compose up -d --build
docker compose ps
The example binds published ports to 127.0.0.1 and explicitly enables unauthenticated local development with AUTH_DISABLED=1. For shared or production access, set AUTH_DISABLED=0, configure a strong API_TOKENS value, set the same token as API_BEARER_TOKEN for the Gradio API backend, and place TLS termination in front of the service.
Check the stack:
curl -sS http://127.0.0.1:8090/health
推奨ツール
別のキーワードを試すか、フィルタを外してください。
インストール
npx skillfish add kdush/openclaw-mirosearch