loopx: Govern Long-Running AI Agent Work

Overview

LoopX is a control plane for agents that need to work toward goals across multiple sessions. It keeps durable state such as tasks, ownership, decisions and evidence, so work can resume after a session ends or an agent runtime changes.

It is intended to work with agent runtimes such as Codex and Claude Code rather than replace them. LoopX governs what should happen next and whether work may continue; the runtime provides the model and tools. This can be useful when a timer or chat history alone is not enough to manage handoffs, review, and decisions that need human attention.

Key Features

• Stores goals, todos, decisions, evidence and run history as durable control state.
• Supports peer task ownership, claims, leases and handoffs between agents.
• Uses quota and continuation checks to govern whether an agent should act again.
• Provides status, review and dashboard surfaces for inspecting progress and pending decisions.
• Integrates with multiple agent hosts, including Codex and Claude Code, with other adapters available.
• Includes optional capabilities for bounded workflows such as issue fixing, research and recurring reports.
• Keeps the kernel as the source of truth when projecting work into dashboards or collaboration surfaces.

Use Cases

• Software teams can track issue or pull-request work across agent sessions while retaining validation evidence and review gates.
• Researchers and experimenters can preserve hypotheses, decisions and results across long-running investigations.
• People managing recurring projects can use goals, schedules and reports to keep work moving between sessions.
• Multi-agent workflows can assign separate tasks to peer agents and make ownership, handoffs and returned results visible.

Getting Started

Install LoopX and its workflow skills:

python3 -m pip install --upgrade loopx && loopx workflow-skills --install

See the README for host-specific setup and usage.

Alternatives

• synara: Synara centers on a desktop workspace for agent sessions, project files, terminals, previews, and Git review, rather than preserving cross-session goals and evidence.
• Coven: Coven focuses on durable, project-scoped coding-agent sessions and authority boundaries across harnesses, rather than broader goal, task, and evidence continuity.
• multica: Multica organizes team work around issues assigned to coding agents and their review, rather than a local-first record of goals and decisions across sessions.
• TaskSync: TaskSync emphasizes task queues, feedback loops, and autonomous coding workflows, rather than durable context and human-visible handoffs across agent sessions.

Considerations

LoopX requires Python 3.11 or later. The README also lists Node.js 22.22.3 or later as a requirement. Host integrations and advanced workflows vary in support and may be optional or experimental. LoopX is not an autonomous production controller: destructive actions, publishing and final ownership remain subject to human authorization.