LazyAgents
TinyWhite1997/LazyAgents
A TUI for manage multiple coding agents like codex, claude code. Vim like operations, with Codex desktop like session/workflow management.
Instalação
npx convez-ai TinyWhite1997/LazyAgents1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
LazyAgents
Keep your coding agents alive across reboots, restarts, and SSH disconnects — without nailing your laptop to a tmux session.
LazyAgents is a long-running local daemon (lad) and a fast TUI (la) that babysits unattended coding agents — Claude Code, OpenAI Codex, sst.dev OpenCode — for you. Sessions survive your terminal closing, cron jobs run them while you sleep, and a built-in per-session git worktree lets you stage and commit what the agent produced.
Install in 5 minutes
Pick one — these are the cargo-dist generated installers; both verify the artifact attestation against sigstore before dropping la and lad onto your $PATH.
Linux / macOS
curl --proto '=https' --tlsv1.2 -LsSf \
https://github.com/TinyWhite1997/LazyAgents/releases/latest/download/lazyagents-installer.sh | sh
Windows (PowerShell)
powershell -ExecutionPolicy ByPass -c "irm https://github.com/TinyWhite1997/LazyAgents/releases/latest/download/lazyagents-installer.ps1 | iex"
From source (any platform)
git clone https://github.com/TinyWhite1997/LazyAgents.git
cd LazyAgents
cargo install --path crates/la-daemon --locked
cargo install --path crates/la-tui --locked
Then launch the TUI — you should land on the Crons tab, and the status bar at the bottom should read daemon @ <socket-path>:
la
The Crons tab is the production reference UI for v1; on a fresh install with no crons yet it renders like this:
┌─ LazyAgents ──────────────────────────────────────────────────────────────┐
│ Projects Sessions [ Crons ] Worktree │
├───────────────────────────────────────────────────────────────────────────┤
│ ENABLED NAME SCHEDULE TZ NEXT FIRE │
│ ─────────────────────────────────────────────────────────────────────────│
│ (no crons yet — press `n` to create one) │
│ │
│ │
├───────────────────────────────────────────────────────────────────────────┤
│ n new e edit ␣ enable/disable d delete r run-now ? help │
│ daemon @ /run/user/1000/lazyagents/lad-1.sock │
└───────────────────────────────────────────────────────────────────────────┘
A full-fidelity screenshot of the same tab once the New-session form lands is tracked as a v1 GA polish item.
Full install path, including service install (systemd / launchd / Windows Scheduled Task) and the macOS configuration fallback chain, is in docs/install.md. Already on LazyAgents and bumping to a new release? See docs/upgrade.md.
Fork / offline / internal mirror?
la --check-updatereads the manifest URL from theLAZYAGENTS_UPDATE_MANIFEST_URLenv var, falling back to GitHub Releases. Point it at your own GitHub-compatible Releases endpoint (response must carrytag_name/html_url/prerelease) when the default host is unreachable. Seedocs/install.md→ Update / uninstall.
Why LazyAgents
You run coding agents long enough that babysitting a single terminal becomes a chore. Claude Code, Codex, and OpenCode are great in the foreground, but the moment you close the lid, hit a flaky VPN, or just want the agent to keep refactoring while you sleep, the session is gone. LazyAgents pins each agent to a local daemon so the session survives the terminal that started it — reattach from anywhere, reboot the laptop, lose the SSH tunnel, the agent keeps running.
It's a real scheduler, not a wrapper around at. Crons carry an IANA timezone per entry, handle DST correctly (docs/adr/0002-cron-dst-fallback-take-first.md), surface catch-up policy when the daemon was down, and budget-cap runaway runs. Nightly claude on the test suite, morning codex on the changelog, hourly opencode audit — write the cron once and the daemon owns the lifecycle.
Local-only, no telemetry. The daemon binds a 0o600 Unix socket (or a SID-restricted Named Pipe on Windows), reads your config from $XDG_CONFIG_HOME/lazyagents/config.toml, persists state in a single SQLite file, and never opens a network listener. Each session gets a dedicated git worktree under $XDG_DATA_HOME/lazyagents/worktrees/, so an agent gone wrong can never overwrite your main checkout. Production diagnostics — Prometheus-format metrics, JSON structured logs, lad doctor health checklist — are documented in docs/observability.md.
Where to go next
| You want to… | Read |
|---|---|
| Get LazyAgents on your machine | docs/install.md |
| Upgrade an existing install (v1.x → v1.y, or plan for v1 → v2) | docs/upgrade.md |
Wire lad doctor + lad metrics into your monitoring | docs/observability.md |
| The full user handbook (chapters per feature, EN + 中文) | docs/book/ |
| Architecture, milestones, design rationale | report/ |
Scope of this release (v1)
LazyAgents v1 is in active development. As of 0.1.1, the three Sessions-tab paths — sidebar list, n New-session, Enter attach to live PTY — are wired end-to-end against the daemon by default (la --demo keeps the in-process fixture for screenshots and design iteration). The release pipeline produces binaries for Linux x86_64/aarch64 (gnu + musl), macOS x86_64/aarch64, and Windows x86_64; Linux is the end-to-end-validated target for v1, with the other platforms tracked as release-blockers for GA. Known platform issues are listed in docs/book/en/src/troubleshooting.md.
License
MIT.