LoopX
A local-first AI agent control plane that preserves goals and review state across sessions, works with Codex and Claude Code, and uses Apache 2.0.
LoopX is a local-first, open-source control plane for people running coding, research or personal AI agents over work that lasts beyond a single session. It keeps goals, decisions and evidence intact when an agent stops, hands off work or resumes elsewhere. Your existing runtime supplies the model and tools; LoopX tracks the work and the decisions that govern its continuation.
Persistent state covers task ownership, run history and recovery as well as the objective itself. You can change runtimes or hosts without rebuilding the goal from chat history. Supported runtimes include Codex, Claude Code, DeepSeek Harness, OpenCode, TraeX and Pi, with support for custom runners.
Human approval gates keep review and publication decisions visible. When one task waits for approval, independent work can continue within its assigned scope. Agent teams share goals while retaining separate ownership and handoffs, and runs record validation and review evidence for the next session.
The Personal Workspace brings goals, tasks, chat and outputs into one view. It's aimed at multi-day projects, issue and PR work, recurring monitoring, and workflows where someone needs to inspect results before allowing the next action.
LoopX uses the Apache 2.0 license and is written in Python. It keeps control state locally; model execution belongs to the connected runtime and provider. The Mac app targets Apple Silicon and requires Python. Background work needs an available host and runtime.
