Skip to content

Backend Selection

Agent Board supports multiple AI backends. Each backend connects to a different AI provider, giving you flexibility in how agents process tasks.

Available Backends

Copilot LM (copilot-lm)

Uses the VS Code Language Model API provided by GitHub Copilot Chat.

Property Value
Requires GitHub Copilot Chat extension
Models Whatever models Copilot provides (Claude, GPT-4o, etc.)
Streaming Yes
Cost Included with Copilot subscription

Recommended

This is the recommended backend for most users as it requires no additional setup beyond the Copilot Chat extension.

Claude CLI (claude-cli)

Uses the Claude CLI tool from Anthropic as a subprocess.

Property Value
Requires Claude CLI installed (brew install claude)
Models Claude models via Anthropic API
Streaming Yes (via subprocess stdout)
Cost Anthropic API pricing

Configuration in board.yaml:

developer:
  cliCommand: claude
  cliArgs:
    - --print

Cline (cline)

Delegates to the Cline VS Code extension for code generation.

Property Value
Requires Cline extension installed
Models Whatever Cline is configured to use
Streaming Via extension API
Cost Depends on Cline's backend

Setting the Default Backend

Via Settings Panel

  1. Open Agent Board sidebar
  2. Click the ⚙️ Settings icon
  3. In the Backend section, click on a backend card
  4. The selected backend shows a green "Available" badge

Via YAML

# .tasks/board.yaml
backends:
  default: copilot-lm

Via VS Code Command Palette

  1. Open Command Palette (Cmd+Shift+P)
  2. Search for "Agent Board: Set Default Backend"

Per-Agent Backend Override

Individual agents can use a different backend than the default:

{
  "id": "agent-developer",
  "name": "Developer",
  "backend": "claude-cli"
}

The resolution order is:

  1. Agent-specific backend field (if set)
  2. Board default backend (from board.yaml)
  3. First available backend (auto-detect)

Backend Availability Detection

When the board loads, it checks which backends are available:

  • Copilot LM — Checks if vscode.lm.selectChatModels() returns models
  • Claude CLI — Checks if the CLI command exists on $PATH
  • Cline — Checks if the Cline extension is installed

Unavailable backends are shown with a gray badge in the settings panel.

Troubleshooting

"No AI backend available"

This error means none of the three backends could be detected. Install at least one:

# Option 1: Install Copilot Chat extension
code --install-extension github.copilot-chat

# Option 2: Install Claude CLI
brew install claude

# Option 3: Install Cline
code --install-extension saoudrizwan.claude-dev

Copilot not detecting models

  • Ensure you're signed into GitHub with an active Copilot subscription
  • Try reloading VS Code: Developer: Reload Window
  • Check Copilot Chat extension is enabled

Claude CLI not found

  • Verify installation: which claude
  • Ensure it's on your $PATH
  • You can set a custom path in board.yaml:
    developer:
      cliCommand: /usr/local/bin/claude