Skip to content

CodeTether Server

Turn AI Agents Into Production Systems.

v1.4.1 Production Release — Agent v1.1.0

CodeTether is battle-tested and running in production at api.codetether.run. The Rust agent now ships with mandatory auth, system-wide audit logging, plugin sandboxing with Ed25519 signing, and Kubernetes self-deployment. See Security Features.

CodeTether Server is a production-ready implementation of the A2A (Agent-to-Agent) Protocol, the open standard from the Linux Foundation that enables AI agents to communicate, collaborate, and solve complex problems together.

  • Get Started in 5 Minutes

    Install CodeTether and run your first multi-agent workflow

    Quick Start

  • Full A2A Protocol Support

    Complete implementation of the A2A specification with extensions

    A2A Protocol

  • Security-First Agent

    Mandatory auth, audit trail, sandboxed plugins, K8s self-deployment

    Security

  • CodeTether Integration

    Bridge AI coding agents to your codebase with session management

    CodeTether

What is CodeTether?

CodeTether Server provides the infrastructure layer for running AI agents in production:

  • Ralph Autonomous Development — Implement entire PRDs autonomously with zero human intervention
  • RLM (Recursive Language Models) — Process arbitrarily large codebases without context limits
  • Zapier Integration — Connect to 5,000+ apps with no-code automation
  • A2A Protocol Native — Full implementation of the A2A specification for agent-to-agent communication
  • Distributed Workers — Run agents across multiple machines with automatic task routing
  • Real-time Streaming — SSE-based live output streaming for long-running agent tasks
  • Session Management — Resume conversations, sync across devices, maintain context
  • Security Features — Mandatory auth, audit trail, plugin sandboxing, K8s self-deployment
  • Enterprise Auth — Keycloak OIDC, API tokens
  • CodeTether Bridge — Native integration with CodeTether AI coding agents
  • Deploy Anywhere — Run on Docker, Kubernetes, or bare metal with ease
  • Email Reply Continuation — Reply to task notification emails to continue conversations with agents

A2A Protocol Implementation

CodeTether implements the complete A2A Protocol specification:

A2A Feature CodeTether Support Reference
Agent Discovery ✅ /.well-known/agent-card.json Spec §5
JSON-RPC 2.0 ✅ Full support Spec §6
Task Management ✅ Create, cancel, status Spec §7
Streaming (SSE) ✅ Real-time artifacts Spec §8
Push Notifications ✅ Webhook callbacks Topics
Multi-turn Conversations ✅ Session continuity Topics

Plus CodeTether extensions:

  • Ralph — Autonomous development loop that implements PRDs with self-healing retry
  • RLM — Recursive Language Models for infinite context processing
  • Zapier — Native OAuth2 integration with triggers, actions, and searches
  • Distributed task coordination — Redis-backed task/state coordination (workers poll the server over HTTP)
  • CodeTether Integration — AI coding agent bridge with codebase registration
  • Monitor UI — Web dashboard for real-time agent observation
  • MCP Tools — Model Context Protocol tool server
  • Email Reply to Continue — Reply directly to task notification emails to keep working with an agent
  • Task Reaper — Automatic stuck task recovery with retry and notification

Quick Start

Option 1: Docker (Server)

# Run the server
docker run -d -p 8000:8000 -p 9000:9000 \
  ghcr.io/rileyseaburg/codetether-server:latest

Option 2: pip (Agent Worker)

The pip package is primarily for installing the agent worker on machines where your code lives:

# Install the worker
pip install codetether

# Configure and start (see installation docs for systemd setup)
codetether-worker --server-url https://api.codetether.run --codebases /path/to/code

Send a Task

curl -X POST http://localhost:8000/v1/a2a \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "method": "message/send",
    "params": {
      "message": {
        "parts": [{"type": "text", "content": "Analyze this codebase"}]
      }
    },
    "id": "1"
  }'

Architecture Overview

At a high level:

  • Clients talk to the API (:8000) for A2A JSON-RPC (/v1/a2a) and REST (/v1/agent/*, /v1/monitor/*).
  • Workers run CodeTether and poll the server for pending tasks over HTTP.
  • Redis provides pub/sub plus shared task/state coordination.
  • The MCP server (:9000) exposes CodeTether capabilities as MCP tools.

For diagrams and deeper detail, see Architecture.

Next Steps