Architecture Overview

Это содержимое пока не доступно на вашем языке.

Pipeline Architecture

┌─────────────────────────────────────────────────────────────┐
│                    OpenPR (Plan)                             │
│  Issues · Boards · Sprints · Governance · MCP · Webhooks     │
└──────────┬──────────────────────────────────┬───────────────┘
           │ Task dispatch                     │ Result callback
           ▼                                   ▲
┌──────────────────────┐              ┌────────────────────┐
│  openpr-webhook      │              │  AI Agent           │
│  Event routing       │── dispatch ──│  Codex / Claude /   │
│  WSS tunnel          │              │  OpenCode           │
└──────────┬───────────┘              └────────┬───────────┘
           │                                   │
           ▼                                   ▼
┌──────────────────────────────────────────────────────────┐
│                      PRX (Think)                          │
│  19 Channels · 14 Providers · Router · Sub-agents         │
│  Self-Evolution · Security Policy · MCP Client            │
└──────────┬──────────┬──────────┬──────────┬──────────────┘
           │          │          │          │
           ▼          ▼          ▼          ▼
      prx-memory  prx-email   Fenfa    WAF + SD
      (Build)     (Build)    (Ship)   (Protect)

Communication Protocols

MCP (Model Context Protocol)

The primary integration protocol. OpenPR exposes 34 MCP tools via HTTP, stdio, and SSE transports. PRX connects as an MCP client.

Webhooks

OpenPR fires 30 event types via HMAC-SHA256 signed webhooks. The openpr-webhook service receives these events and dispatches work to AI agents.

WSS Tunnel

For agents behind NAT/firewalls, openpr-webhook supports an outbound WebSocket tunnel to the OpenPR control plane. Tasks are received, acknowledged, executed, and results returned through the tunnel.

Data Flow

Issue → Fix → Deploy

Issue created in OpenPR (manually or via API)
Bot task assigned — OpenPR fires issue.created webhook with bot context
Webhook dispatcher receives the event, matches it to an agent configuration
CLI executor launches the coding agent (e.g., claude-code) with a templated prompt
Agent works — reads code, makes changes, runs tests
Agent uses prx-memory via MCP to recall past patterns and store new learnings
Result callback — agent posts results back to OpenPR via MCP/API
State transition — issue moves from in_progress to done
CI builds — standard CI/CD pipeline produces artifacts
Fenfa distributes — build artifacts uploaded via API, distributed to all platforms

Security Loop

WAF monitors incoming HTTP traffic through 17 detection phases
SD scans files on endpoints with hash matching, YARA rules, and ML inference
Threats detected → automated response (block, quarantine, remediation)
Notifications sent via webhook, email, Telegram
Future: Security events feed back into OpenPR as issues for AI-driven response

Deployment Topology

Minimal (Single Machine)

┌─────────────────────────┐
│  Server                  │
│  ├── OpenPR (Docker)     │
│  ├── PRX (binary)        │
│  ├── prx-memory (binary) │
│  └── PostgreSQL          │
└─────────────────────────┘

Production

┌──────────┐  ┌──────────┐  ┌──────────┐
│ OpenPR   │  │ PRX      │  │ Fenfa    │
│ + Caddy  │  │ daemon   │  │ + S3/R2  │
└────┬─────┘  └────┬─────┘  └──────────┘
     │             │
     ▼             ▼
┌──────────┐  ┌──────────┐
│ WAF      │  │ SD       │
│ cluster  │  │ daemon   │
└──────────┘  └──────────┘