EchoBird Documentation

EchoBird is an AI deployment and management agent designed to simplify the installation and configuration of AI tools and local models. It uses a central Model Nexus (模型中心) to allow configuration in one place to apply across multiple scenarios.

Core Scenarios

• Installation & Repair: Use AI to install and fix mainstream AI tools (e.g., Claude Code, OpenClaw, Hermes Agent) locally or remotely.
• One-click Local LLM: Run local models using built-in engines like vLLM, SGLang, or llama.cpp by selecting a quantization version.
• My AI Projects: Unified integration and management for your own Vibe Coding applications or games.
• App Management: One-click launch and management for all AI and Agent-related applications.

Model Nexus (模型中心)

Acts as a unified data hub for OpenAI, Anthropic, local LLMs, and API Routers. Configuring a provider here enables it to be used across all supported tools without manually editing TOML or JSON configuration files.

EchoBird uses a Rust-native Codex proxy to bridge the gap between the Codex CLI/Desktop and various LLM providers (DeepSeek, Moonshot, etc.).

Workflow:

1. Configuration: EchoBird writes a canonical ~/.codex/config.toml so the Codex CLI points to the local proxy (base_url = "http://127.0.0.1:53682/v1").
2. Proxy Service: The Rust proxy binds to 127.0.0.1:53682 and serves POST /v1/responses. It reads ~/.echobird/codex.json on every request to allow real-time model and API key switches without a restart.
3. Translation: The proxy translates incoming 'Responses' format requests into 'Chat Completions' requests for the upstream provider, then translates the streaming response back to 'Responses' SSE format.
4. Model Spoofing: It rewrites Codex-specific model IDs (like gpt-5.4) to the actual upstream model ID and mirrors it back to maintain Codex's internal bookkeeping.

Before installing any agent (OpenClaw, Hermes Agent, Claude Code, Codex, etc.), you must perform these three checks in order:

1. Platform Compatibility Check

Detect the OS and architecture using uname -s && uname -m and verify against supported platforms:

2. Node.js Version Check

For npm-based agents, verify Node.js meets the minimum version:

• OpenClaw: >= 22.14.0 (Critical)
• Codex (CLI): >= 18.0.0

To install/upgrade Node.js via nvm (Linux/macOS):

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.3/install.sh | bash
source ~/.nvm/nvm.sh
nvm install 22
nvm use 22
node --version

To install Node.js on Windows (PowerShell):

Invoke-WebRequest -Uri "https://nodejs.org/dist/v22.14.0/node-v22.14.0-win-x64.zip" -OutFile "$env:TEMP\node.zip"
Expand-Archive -Path "$env:TEMP\node.zip" -DestinationPath "C:\nodejs" -Force
$nodePath = "C:\nodejs\node-v22.14.0-win-x64"
[Environment]::SetEnvironmentVariable("PATH", "$nodePath;" + [Environment]::GetEnvironmentVariable("PATH", "Machine"), "Machine")
& "$nodePath\node.exe" --version

Note: On Windows, you must close and reopen terminals for PATH changes to take effect.

3. Download Speed Test

Run this on the remote server to ensure the server can reach the source quickly:

For npm-based agents:

curl -o /dev/null -s -w "%{time_total}" https://registry.npmjs.org/openclaw/latest 2>/dev/null

For GitHub-hosted CLI installers:

curl -o /dev/null -s -w "%{time_total}" https://raw.githubusercontent.com 2>/dev/null

If the response time is > 5 seconds or times out, do not proceed. Ask the user for a VPN/proxy, a local installer file, or alternative mirrors.

When installing desktop apps (e.g., Claude Desktop, ChatGPT, Gemini Desktop), follow these rules:

1. Target Local Machine: Desktop apps must be installed on the user's local machine. If connected to a remote server, switch to the local server before proceeding.
2. Follow install_flow.agent_steps: Execute the specific steps provided in the tool's installation JSON.
3. Manual Interaction: Do NOT attempt to automate GUI wizards (no silent-install flags like /S or /VERYSILENT unless explicitly specified).
   • Tell the user the exact download path.
   • Open the installer for them.
   • Instruct them to click through the wizard (Next → Next → Install).
4. Use Package Managers: On Windows, prefer winget for faster, silent installation:
   • Example for ChatGPT: winget install --id 9PLM9XGG6VKS
5. Handling Windows .ps1 Installers: To prevent the agent from hanging while waiting for a GUI wizard, wrap the command in a detached Start-Process:

Start-Process powershell -ArgumentList "-NoProfile","-Command","irm <URL> | iex"

6. Platform Check: Only proceed if the user's OS is listed in the tool's platforms configuration. If unsupported, suggest the web alternative.

Use these instructions to guide an agent through a comprehensive security audit on a remote server. The audit covers seven key areas to identify brute force attacks, malware, network anomalies, and configuration weaknesses.

Audit Checklist

1. SSH Brute Force: Count failed login attempts in auth logs and identify top attacker IPs. A count > 100 indicates an active attack.
2. Active Users: Check currently logged-in users and flag unexpected sessions.
3. Malware & Crypto Miners:
   • Search processes for: xmrig, kdevtmpfsi, kinsing, minergate, cpuminer, ccminer, ethminer, xmr-stak, cryptonight, stratum, coinhive.
   • Check for high CPU usage.
   • Scan /tmp, /var/tmp, and /dev/shm for hidden executables.
4. Network: List listening ports, check established outbound connections for unusual IPs, and look for reverse shells.
5. SSH Security Config: Check for default port (22), root login enabled, password authentication enabled, and unexpected authorized keys.
6. Cron Jobs: List user cron jobs and /etc/cron.d/ files. Flag entries using curl, wget, bash, or python with remote URLs.
7. System Integrity: List files modified in /etc, /usr/bin, or /usr/sbin in the last 24 hours. Check kernel version, uptime, and fail2ban status.