Agent Queue

Agent Queue

**A self-improving orchestration platform for AI coding agents.**

Agent Queue manages task queues across multiple projects, coordinates multi-agent workflows through executable playbooks, accumulates knowledge via a 4-tier memory architecture, and continuously improves through automated reflection. Every completed task feeds insights back into the system — the longer it runs, the better it gets.

Manage everything from Discord, your terminal, or any MCP-compatible client. Queue up a week's worth of tasks, come back to completed PRs and a system that knows more about your codebase than it did yesterday.

Key Capabilities

Self-Improvement Loop

The core value proposition: **the system gets better with use.**

• Reflection engine. Post-task review extracts generalizable insights — what worked, what failed, what to remember. Configurable depth tiers (deep/standard/light) with circuit breaker protection.
• Continuous learning. Every task leaves the system better prepared for the next one. Error patterns, successful strategies, and project conventions accumulate in scoped memory.
• Knowledge consolidation. Periodic consolidation distills raw task memories into structured knowledge bases, project factsheets, and agent-type wisdom.
• Autonomous operation. The improvement loop requires no manual intervention — reflection playbooks run automatically, write insights, and feed them to future agents.

Orchestration & Scheduling

• Multi-agent, multi-project. Run parallel Claude Code agents across all your projects, each in isolated workspaces with your existing environment.
• Proportional scheduling. Deficit-based fair-share algorithm distributes agent