agent-nexus

One Command. All Your AI Agents United.

Shared memory + Telegram remote control for Claude Code, Codex, and Gemini CLI -- installed in 60 seconds.

English | 简体中文

---

The Problem

You have Claude Code, Codex, maybe Gemini CLI. They're powerful individually. But right now:

• They don't share memory. Tell Claude your coding style, Codex has no idea.
• They can't be controlled remotely. Step away from your desk? Your agents are unreachable.
• Setting up shared infrastructure is painful. Clone this, configure that, edit three different config files...

You could spend an afternoon installing RecallNest for shared memory, telegram-ai-bridge for remote control, and manually wiring MCP configs for each tool.

Or:

npm i -g @aliceljy/agent-nexus && agent-nexus init

What Happens When You Run init

$ agent-nexus init

  Detecting environment...
  ✅ Bun
  ✅ Claude Code
  ✅ Codex CLI
  ⬚ Gemini CLI (not installed, skipping)

  📝 Credentials

  Your Telegram User ID: 123456
  Jina API Key: jina_****
  HTTPS Proxy: (skipped)

  🤖 Bot Tokens (each backend needs its own @BotFather token)

  claude Bot Token: 111:AAA****
  codex Bot Token: 222:BBB****

  🔗 CC ↔ Codex Communication

  Choose [3]: 3  (Both)

  🔌 Configuring MCP...

  📋 Backed up ~/.claude.json
  ✅ Claude Code MCP configured
  ✅ Codex MCP configured
  ✅ Bridge config generated

  🔗 CC → Codex: Install the official plugin in Claude Code:
     claude /install-plugin codex@openai-codex

  🔗 CC → Codex via RecallNest: ✅ Already configured
  🔗 Codex → CC: claude -p "your prompt" (no extra config needed)

  ✅ agent-nexus setup complete!
  Run: agent-nexus start

That's it. Your Claude Code and Codex now share one brain. Your phone is now the remote.

Before / After

	Before agent-nexus	After agent-nexus
Memory	Each agent starts from zero every session	All agents share persistent memory via RecallNest
Remote control	Must be at your desk	Full control from Telegram on your phone
Setup time	30+ min (clone, configure, debug MCP paths...)	60 seconds
Config files to edit	3-5 (per tool, per service)	0 (auto-injected)
Processes to manage	Start each service manually	agent-nexus start / stop

Install

npm i -g @aliceljy/agent-nexus

Requires Bun >= 1.3 and at least one of: Claude Code, Codex CLI, or Gemini CLI.

Usage

agent-nexus init      # Interactive wizard -- detects tools, collects keys, writes all configs
agent-nexus start     # Start RecallNest + Telegram bridge in background
agent-nexus stop      # Stop all services
agent-nexus status    # Health check with memory stats

How It Works

                    agent-nexus init
                         |
          +--------------+--------------+
          |              |              |
     ~/.claude.json  ~/.codex/     ~/.gemini/
     (MCP injected)  config.toml   settings.json
          |              |              |
          v              v              v
     Claude Code     Codex CLI    Gemini CLI
          |              |              |
          +------+-------+------+------+
                 |              |
            RecallNest    telegram-ai-bridge
            (memory)      (remote control)
                 |              |
            LanceDB         Telegram
          (your brain)    (your phone)

agent-nexus doesn't replace anything. It's the glue that makes your existing tools work as a team.

Cross-Agent Communication

Beyond shared memory, your agents can call each other directly.

Claude Code → Codex

During agent-nexus init, choose one or both:

Method	How It Works	Best For
Official Plugin	codex@openai-codex plugin for Claude Code	Direct task delegation, rescue debugging
Shared MCP	Both agents read/write RecallNest	Context sharing, async handoffs

Official Plugin -- Claude Code can spawn Codex as a subagent for parallel tasks, second opinions, or rescue debugging:

claude /install-plugin codex@openai-codex

Shared MCP -- RecallNest is auto-configured as an MCP server for both tools. Claude Code writes context, Codex reads it (and vice versa). No extra setup needed.

Codex → Claude Code

Codex can call Claude Code directly via CLI -- no plugin or MCP required:

claude -p "analyze this function for security issues"

If you hit TTY issues in non-interactive environments: script -q /dev/null claude -p "prompt"

Group Chat (Multi-Bot in One Telegram Group)

Put your CC bot and Codex bot in the same Telegram group and they'll share context automatically. During agent-nexus init, if 2+ backends are enabled, you'll be asked:

👥 Group Chat

Put multiple bots in the same Telegram group? [y/N]: y

Shared context backend:
  1. SQLite (local, single machine)
  2. Redis (recommended for multi-bot groups)
  Choose [2]: 2

Redis URL [redis://localhost:6379]:

Why Redis? Each bot runs as a separate process. SQLite works for single-machine setups, but Redis gives you:

• Real-time cross-process context sharing (no file locks)
• TTL-based auto-cleanup (no stale messages piling up)
• Ready for multi-machine deployment if you scale later

Telegram bots can't see each other's messages natively. The shared context layer works around this by syncing messages through the backend store.

Why Not A2A?

The telegram-ai-bridge project includes an A2A bus for multi-bot auto-broadcast in Telegram group chats. agent-nexus supports A2A but defaults to off:

• Most users don't need it. 1:1 private chats are the common case, and bots talk to each other directly via MCP/CLI.
• Group war rooms do. Want multiple bots chiming in automatically in the same group? Set a2aEnabled: true in bridge-config.json to turn it on.

What Gets Installed

Component	What It Does	Already Have It?
RecallNest	Shared memory across all agents (MCP + HTTP API)	Keeps yours, adds if missing
telegram-ai-bridge	Telegram remote control for AI agents	Keeps yours, adds if missing
MCP configs	RecallNest auto-registered in each tool	Non-destructive merge (backup first)

Prerequisites

• Bun >= 1.3
• At least one AI CLI tool:
  • Claude Code
  • Codex CLI
  • Gemini CLI
• A Telegram bot token (free, from @BotFather)
• A Jina API key (free tier available at jina.ai)

Config

Configuration lives in ~/.agent-nexus/:

config.json -- nexus-only settings:

{
  "jinaApiKey": "jina_...",
  "crossAgent": { "ccToCodex": "both" }
}

bridge-config.json -- auto-generated from telegram-ai-bridge's config API, no manual sync needed:

{
  "shared": { "ownerTelegramId": "123456", "sharedContextBackend": "redis", ... },
  "backends": {
    "claude": { "enabled": true, "telegramBotToken": "111:AAA..." },
    "codex": { "enabled": true, "telegramBotToken": "222:BBB..." },
    "gemini": { "enabled": false }
  }
}

Each backend needs its own bot token from @BotFather.

Ecosystem

agent-nexus is the installer for a larger ecosystem of AI agent tools:

Project	Role	Stars
RecallNest	Shared memory layer (LanceDB + hybrid retrieval)
telegram-ai-bridge	Telegram remote control + multi-agent orchestration
Claude Code Studio	Multi-session collaboration platform for Claude Code

Want the full workflow? cc-cabin is the scaffold that ties these modules together — hooks, rules, workflow methodology, and more. agent-nexus gives you memory + remote in 60 seconds; cc-cabin gives you the complete Claude Code workflow.

License

MIT