Weco DocsUse your agent in the dashboard
Install the Weco skill and bridge Claude Code sessions to the dashboard.
setup
Install the Weco skill into AI coding assistants so they can drive Weco for you (see the Skills guide):
# Interactive picker (defaults to all supported tools)
weco setup
# Or target a specific tool
weco setup claude-code
weco setup cursor
weco setup codex
weco setup openclaw
# Install for every supported tool at once
weco setup allThe skill is installed into the tool's skills folder in your home directory (e.g. ~/.claude/skills/weco for Claude Code). Use --local <path> to install from a local weco-skill checkout instead of downloading (for development).
start
weco start claude launches Claude Code wired to the Weco dashboard, so a coding session streams live to the web and you can drive or watch it from there. The agent's messages, tool calls, and reasoning appear in the dashboard, and tool-approval prompts can be answered from either the terminal or the dashboard.
# Launch Claude Code with a local TUI bridged to the dashboard
weco start claude
# Seed the first turn with a prompt
weco start claude -p "Use Weco to make forward() faster in model.py"
# Headless: no local UI, stream only to the dashboard (good for background sessions)
weco start claude --headless --allow-tools -p "Optimize the API handler for latency"
# Bill LLM usage through your Weco credit wallet instead of your local Claude auth
weco start claude --billing weco
# Forward extra flags to the underlying claude CLI (after a -- separator)
weco start claude -- --model claude-opus-4-8weco start claude runs the Claude Code CLI under the hood, so it must be installed and on your PATH (install guide). You also need to be logged in to Weco (weco login).
Start Command Options
| Argument | Description | Default |
|---|---|---|
-p, --prompt | Seed the first turn with this prompt instead of waiting for input. Required to make --headless do anything immediately; also works in the TUI. | None |
--headless | Run with no local TUI — stream only to the dashboard, printing key lifecycle lines to the console. Use when launching in the background (no terminal to draw into). Pair with --allow-tools and --prompt. | False |
--allow-tools | Auto-approve all Claude Code tool calls (passes --dangerously-skip-permissions). Bash, Write, Edit, etc. run without prompting. Use only when you trust the agent. | False |
--effort | Thinking effort level (low, medium, high, xhigh, max). Raises the thinking-token budget so Claude returns reasoning that streams into the dashboard's Reasoning section. Omit to inherit Claude's own default. | None |
--billing | Where LLM calls are billed. claude (default) uses your local Claude auth (claude login or ANTHROPIC_API_KEY). weco routes through Weco's LLM proxy and deducts from your Weco credit wallet. | claude |
claude_args | Extra arguments forwarded to the underlying claude CLI. Prefix with -- to separate them from Weco's own flags. | - |