PRX

PRX is a high-performance, self-evolving AI agent runtime written in Rust. It connects large language models to 19 messaging platforms, provides 46+ built-in tools, supports WASM plugin extensions, and autonomously improves its own behavior through a 3-layer self-evolution system.

PRX is designed for developers and teams who need a single, unified agent that works across every messaging platform they use -- from Telegram and Discord to Slack, WhatsApp, Signal, iMessage, DingTalk, Lark, and more -- while maintaining production-grade security, observability, and reliability.

Why PRX?

Most AI agent frameworks focus on a single integration point or require extensive glue code to connect different services. PRX takes a different approach:

• One binary, every channel. A single prx binary connects to all 19 messaging platforms simultaneously. No separate bots, no microservice sprawl.
• Self-evolving. PRX autonomously refines its memory, prompts, and strategies based on interaction feedback -- with safety rollback at every layer.
• Rust-first performance. 177K lines of Rust deliver low latency, minimal memory footprint, and zero-GC pauses. The daemon runs comfortably on a Raspberry Pi.
• Extensible by design. WASM plugins, MCP tool integration, and a trait-based architecture make PRX easy to extend without forking.

Key Features

• 19 Messaging Channels -- Telegram, Discord, Slack, WhatsApp, Signal, iMessage, Matrix, Email, Lark, DingTalk, QQ, IRC, Mattermost, Nextcloud Talk, LINQ, CLI, and more.

• 9 LLM Providers -- Anthropic Claude, OpenAI, Google Gemini, GitHub Copilot, Ollama, AWS Bedrock, GLM (Zhipu), OpenAI Codex, OpenRouter, plus any OpenAI-compatible endpoint.

• 46+ Built-in Tools -- Shell execution, file I/O, browser automation, web search, HTTP requests, git operations, memory management, cron scheduling, MCP integration, sub-agents, and more.

• 3-Layer Self-Evolution -- L1 memory evolution, L2 prompt evolution, L3 strategy evolution -- each with safety bounds and automatic rollback.

• WASM Plugin System -- Extend PRX with WebAssembly components across 6 plugin worlds: tool, middleware, hook, cron, provider, and storage. Full PDK with 47 host functions.

• LLM Router -- Intelligent model selection via heuristic scoring (capability, Elo, cost, latency), KNN semantic routing, and Automix confidence-based escalation.

• Production Security -- 4-level autonomy control, policy engine, sandbox isolation (Docker/Firejail/Bubblewrap/Landlock), ChaCha20 secret store, pairing authentication.

• Observability -- OpenTelemetry tracing, Prometheus metrics, structured logging, and a built-in web console.

Architecture

graph TB
    subgraph Channels["Messaging Channels"]
        TG[Telegram]
        DC[Discord]
        SL[Slack]
        WA[WhatsApp]
        SIG[Signal]
        iM[iMessage]
        MX[Matrix]
        EM[Email]
        LK[Lark]
        DT[DingTalk]
        MORE[+ 9 more]
    end

    subgraph Core["PRX Core"]
        RT[Agent Runtime]
        LOOP[Agent Loop]
        SEC[Security Policy]
        CFG[Config Hot-Reload]
    end

    subgraph Providers["LLM Providers"]
        AN[Anthropic]
        OA[OpenAI]
        GE[Gemini]
        OL[Ollama]
        BR[Bedrock]
        CP[Copilot]
        OR[OpenRouter]
    end

    subgraph Router["LLM Router"]
        HR[Heuristic]
        KNN[KNN Semantic]
        AM[Automix]
    end

    subgraph Tools["46+ Tools"]
        SH[Shell]
        FS[File I/O]
        BW[Browser]
        WS[Web Search]
        GT[Git]
        MCP[MCP]
        SUB[Sub-agents]
        CRON[Cron]
    end

    subgraph Memory["Memory System"]
        MD[Markdown]
        SQ[SQLite]
        PG[PostgreSQL]
        EMB[Embeddings]
    end

    subgraph Evolution["Self-Evolution"]
        L1[L1: Memory]
        L2[L2: Prompt]
        L3[L3: Strategy]
    end

    subgraph Plugins["WASM Plugins"]
        PW[Plugin Worlds]
        PDK[PDK]
    end

    GW[Gateway HTTP/WS]

    Channels --> RT
    RT --> LOOP
    LOOP --> Router
    Router --> Providers
    LOOP --> Tools
    LOOP --> Memory
    LOOP --> Evolution
    LOOP --> Plugins
    RT --> SEC
    RT --> CFG
    GW --> RT

Quick Install

curl -fsSL https://openprx.dev/install.sh | bash

Or install via Cargo:

cargo install openprx

Then run the onboarding wizard:

prx onboard

See the Installation Guide for all methods including Docker and building from source.

Documentation Sections

Section	Description
Installation	Install PRX on Linux, macOS, or Windows WSL2
Quick Start	Get PRX running in 5 minutes
Onboarding Wizard	Configure your LLM provider and initial settings
Channels	Connect to Telegram, Discord, Slack, and 16 more platforms
Providers	Configure Anthropic, OpenAI, Gemini, Ollama, and more
Tools	46+ built-in tools for shell, browser, git, memory, and more
Self-Evolution	L1/L2/L3 autonomous improvement system
Plugins (WASM)	Extend PRX with WebAssembly components
Configuration	Full config reference and hot-reload
Security	Policy engine, sandbox, secrets, threat model
CLI Reference	Complete command reference for the prx binary

Project Info

• License: MIT OR Apache-2.0
• Language: Rust (2024 edition)
• Repository: github.com/openprx/prx
• Minimum Rust: 1.92.0