asynq-org

Asynq Agentd — AI skill for Claude Code

AI community

An open-source daemon that keeps your AI coding agents (Claude Code, Codex, OpenCode) running 24/7 on any always-on machine.

How to install Asynq Agentd

This entry records only its repository, not the path inside it, so there is no exact command to give. Open asynq-org/asynq-agentd and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Asynq Agentd does

An open-source daemon that keeps your AI coding agents (Claude Code, Codex, OpenCode) running 24/7 on any always-on machine. Task queue, approval gates, structured activity logs, session continuity, and a local API — so you can monitor, steer, and approve agent work from your phone or any browser. The control plane for your AI workforce.

Alternatives in AI

Dsh Auto Review — Second-model AI auto-review for DeepSeek Harness approval requests: a read-only reviewer subagent returns stru 155 ★
AppleAI — Apple AI is a lightweight, privacy‑minded macOS menu bar app that keeps your favorite AI assistants one keystr 145 ★
Token Proxy — Local AI API gateway for OpenAI / Gemini / Anthropic 92 ★

README

asynq-agentd

`asynq-agentd` is the open-source daemon and local CLI behind the broader Asynq Buddy product direction.

Status

`asynq-agentd` is in early public beta.

The repo is already useful if you want to:

run a local daemon for Claude Code, Codex, and Cursor
queue work, approvals, and recurring tasks
inspect recent work and activity
use a small operator CLI while the full Buddy UI is still being built separately

What is still intentionally beta:

installer polish and hosted one-line install
OpenCode adapter implementation
deeper recovery semantics
stronger release packaging and release automation

What is implemented

Native TypeScript runtime using Node.js only
SQLite-backed persistence via node:sqlite
In-process scheduler with task dependencies and bounded parallelism
Session, task, approval, activity, config, and stats REST endpoints
Dashboard-oriented REST endpoints for overview, attention-required, and continue-working cards
Background summary enrichment for dashboard cards with provider auto-selection (claude -> codex -> heuristic)
Pluggable adapter layer with real Codex and Claude Code CLI adapters plus mock fallbacks
Bearer-token auth with local token persistence in .asynq-agentd/auth.json
Project-level .asynq-agentd.yaml inheritance for task context
Project-level approval/model routing overrides via .asynq-agentd.yaml
Recurring tasks via cron expressions
Codex task execution via codex exec --json with live activity import
Claude Code task execution via claude -p --output-format stream-json
Basic recovery for Codex-backed running tasks after daemon restart
Basic recovery for Claude/Codex-backed running tasks after daemon restart
PID-aware runtime reconciliation to avoid duplicating already-running CLI sessions after restart
Resume-confidence recovery that pauses interrupted CLI tasks when no safe external session id is available
First runtime approval interception hooks for