Agent launch guide
Agent Launch opens a coding agent CLI in your own terminal — credentials materialized, model routing configured, project integrations installed — instead of hosting it inside the daemon. Use it when you want the full...
Agent Launch opens a coding agent CLI in your own terminal — credentials materialized, model routing configured, project integrations installed — instead of hosting it inside the daemon. Use it when you want the full native CLI experience with VibeAround's profile and continuity benefits. (Curious what exactly gets injected and how? See the launch subsystem internals.)
Launching from the desktop app
The Launch screen asks for three choices:
- Agent — any enabled agent, including the desktop-app targets (
claude-desktop,codex-desktop) which open the vendor's GUI app instead of a CLI. - Workspace — the directory the agent starts in.
- Model profile —
direct(agent's own vendor login) or one of your provider profiles (see Model profiles guide).
VibeAround renders the launch — environment variables, per-agent config overlays, bridge URLs when the profile needs them — and opens your terminal app (Terminal.app or iTerm2 on macOS, PowerShell on Windows, xdg-terminal-exec/common terminals on Linux). Terminal preference is configurable.
Launching from the CLI
Saved launch profiles live in ~/.vibearound/launch/profiles/<name>.json:
va launch --profile codex # by name
va launch --profile-path ./my.json # by file
va launch --profile codex --dry-run # validate and print the plan, launch nothingva launch execs the bundled native va-launch binary — a standalone launcher that also works by itself (/path/to/va-launch --profile <name>), so scripted and CI-adjacent launches do not need the full CLI.
A launch profile is a small JSON document (schema version 1): agent, workspace, terminal choice, command/executable override, env, args, window label. Unknown fields are rejected to catch handing it the wrong kind of JSON.
What happens at launch
- Validation. Workspace exists, agent executable resolves (explicit path →
~/.vibearound/agents.json→ PATH scan, cached after first discovery). - Project integrations. If the local daemon is running, project-scoped MCP config and skill files for the agent are installed into the workspace (honoring
integrations.mcp_auto_installandintegrations.skill_auto_install). If the daemon is not running, VibeAround-managed project integrations are removed instead — stale config must not point the agent at a dead MCP server. - Terminal spawn. The agent starts in your terminal with the rendered environment. For bridged profiles, its model traffic flows through
127.0.0.1:12358— so keep the daemon running for the session's lifetime.
Sessions from launches
Launched CLIs create their own native sessions, which VibeAround discovers:
va launch sessions # resumable sessions across agents/workspaces
va launch archive --agent claude <id> # hide one from pickers
va launch unarchive --agent claude <id>These sessions appear in the desktop/dashboard resume pickers and can be attached to a chat with /session --switch <id> or handed over with the in-CLI handover tool + /pickup <code>. That is the bridge between "working in a terminal" and "continuing from a phone" — see Session lifecycle.
Launch vs hosted: which to use
| Launch (your terminal) | Hosted (IM / web chat) | |
|---|---|---|
| UI | The agent's own full TUI | Chat bubbles + permission cards |
| Process owner | Your terminal | The daemon (idle-shutdown managed) |
| Model routing | Profile-rendered config | Same profiles, same bridge |
| Continuity | Sessions discoverable, handover via code | Sessions tracked on the thread automatically |
| Survives daemon stop | CLI keeps running; bridged model calls fail until daemon returns | No (agent is daemon-hosted) |
Troubleshooting
| Symptom | Fix |
|---|---|
| "executable not found" | Install the agent CLI, or set an explicit executablePath in the launch profile; delete a stale entry in ~/.vibearound/agents.json to force a re-scan |
| Terminal opens then closes instantly | Run with --dry-run to see the plan; try the command it prints manually to surface the real error |
| Agent starts but model calls fail | Daemon not running (bridged profile), or profile key invalid — check va status and the profile |
| Linux: nothing opens | No supported terminal found; install one of the common terminals or set the terminal preference |
Source anchors: src/launcher/ (va-launch: validation, integrations, spawn), ~/.vibearound/launch/profiles/ schema (launch profile JSON v1), src/core/src/agent/launch.rs + src/core/src/profiles/bridge_launch.rs (profile rendering), src/core/src/launch_sessions/ (session discovery), src/cli/src/args.rs (va launch commands).
Last verified: v0.7.11
Claude Code Provider Switcher
The scenario: you want to run Claude Code, but through a provider and key you chose — DeepSeek, Kimi, GLM, OpenRouter, or a self hosted compatible endpoint — without editing environment variables before every session...
IM usage
How to drive a coding agent from a chat window, and the complete slash command reference. This page applies to every channel — Telegram, Slack, Feishu, Discord, WeChat, DingTalk, WeCom, QQ Bot — plus the built in web...