Synaplan
开源 AI 软件平台,面向企业:自托管 AI 代理,符合 GDPR 的聊天机器人小部件,RAG 知识库,WhatsApp 与 Outlook AI,MCP,多模型路由。支持 Docker、Kubernetes 或在 Germ 托管。
AI 翻译
尚未检测到更新。GitHub 发布版本后会显示在这里。
正在加载完整 README。 以下为已保存的摘录。
Government case-intake demo: an agentic AI processes incoming service
applications end-to-end with a human checkpoint. Built for a 2-3 minute
screen recording.
Open http://localhost:5173. No API key required - the demo runs on
deterministic canned outputs out of the box (badge shows LLM FALLBACK).
Set ANTHROPIC_API_KEY in .env for live Claude classification,
extraction and drafting (badge shows LLM LIVE).查看完整 README ↗Government case-intake demo: an agentic AI processes incoming service applications end-to-end with a human checkpoint. Built for a 2-3 minute screen recording.
npm install
npm run dev
Open http://localhost:5173. No API key required - the demo runs on
deterministic canned outputs out of the box (badge shows LLM FALLBACK).
Set ANTHROPIC_API_KEY in .env for live Claude classification,
extraction and drafting (badge shows LLM LIVE).
Three modes, via AGENT_LLM_MODE and AGENT_LLM_BUDGET:
live (default): calls Claude. AGENT_LLM_BUDGET caps live calls per
server instance (default 60, ~30 full runs) - bounds spend even if the
public URL leaks.replay: never calls the API. Serves seed/live-cache.json - real
Claude outputs captured once with npm run capture, committed to the
repo. Badge shows LLM REPLAY. Misses fall back to canned data.The public deployment is configured with AGENT_LLM_MODE=replay.