loopx

LoopX 2.0

Turn long-horizon goals into verifiable results.

A local-first control plane and research workbench for agents and teams. Keep goals, decisions, ownership and evidence across sessions. Work with the runtimes you already use: Codex, Claude Code, DeepSeek Harness and others.

Get started · Why LoopX · See it in action · Evidence · Contribute · Docs · 简体中文

Start in two steps

With Python 3.11+ and Node.js 22.22.3+ (24 LTS recommended), install once:

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

Restart your Agent host, open your project, then give Codex a concrete task:

$loopx speed up this project's slowest workflow; compare approaches and verify the gains

Claude Code uses /loopx . Other hosts, install options and upgrades ↓

Why LoopX

An agent can finish a task in one session. Long-running work is harder: objectives change, owner decisions appear, evidence goes stale, and agents hand work to peers. A timer can keep a loop running without making it useful.

• Continue without starting over. Keep the Goal, accepted requirements,

decisions and evidence across sessions and restarts.

• Coordinate without losing ownership. Connect delegated work, dependent

artifacts, review and returned results while keeping responsibility explicit.

• Judge progress by what changed. Inspect deliverables, supporting and

refuting evidence, costs and the next decision—not just activity.

Your Agent runtime supplies the model and tools. LoopX carries durable control state for long-horizon agents and personal agent teams across engineering, research and operations. Human authorization still governs protected actions.

What you want to do · Start here Keep a coding or research agent working across sessions · Install and start Manage personal projects, schedules and decisions in one place · Personal Agent Workspace Let agents collaborate and deliver verifiable results · Agent collaboration guide

LoopX keeps the durable control state in one compact layer:

Agent runtimes execute the work. LoopX governs the state that lets engineering, research, discovery, and operations loops continue across runs. It is not another agent framework or a provider-specific orchestration runtime.

How the control plane works and where it fits

A useful mental model is an agent-native Kanban for long-running work. Cards carry identity, authority, evidence, and continuation. Moves are validated operators such as claim, gate, monitor, and writeback. The board is a projection; LoopX state remains the source of truth.

Registered agents are peers. Claims, leases, task boundaries, capabilities, and typed continuation decide who acts next; no durable leader identity is required.

LoopX is useful when you run:

• multi-day engineering, research, benchmark, or experiment objectives;
• issue and PR loops that must preserve scope, evidence, and review state;
• recurring heartbeat or monitor work;
• projects with owner, safety, publication, or private-data gates;
• peer-agent teams where ownership, leases, and handoff matter;
• creator, research, or operations workflows whose progress must remain

legible to a non-engineering operator.

LoopX is not an autonomous production controller. Dangerous permissions, publishing, production writes, and final ownership stay with the human.

See LoopX in Action

Watch the research-workbench film · 86s · transcripts, subtitles and sources. Recorded App excerpts, synthetic examples and illustrations are labelled.

Research that can continue

Open a Goal to inspect its work, responsible Agents and recorded Explore history. Use Deep analysis to inspect a registered trajectory, its citations, native outcome readback and separately measured costs. An Agent prepares the source; opening a page or reading a result does not start a model call. Explicit analysis uses the configured host and account.

Keep supported, refuted and unknown findings visible. Inspect why an approach changed or continued, then carry the evidence into the next attempt.

Meet the Personal Agent Workspace

Keep long-horizon goals in one local-first workspace. Goals, attention, conversations, tasks, files, schedules, and recovery stay durable across days, restarts, and harnesses. Reopen a project, inspect the previous turn’s state and evidence, and continue the next permitted action.

LoopX brings these long-horizon control states into the Personal Workspace. It gives you one place to:

• see what needs you, what is running, what is being watched, and what is

scheduled or stopped;

• configure Goal capabilities, distinguish machine defaults from Goal overrides,

and preview changes before applying them;

• steer a live turn, queue a message, or use the async inbox from a connected

Lark conversation with explicit Goal/Agent/session routing;

• inspect deliverable files, periodic reports, and their supporting evidence;
• continue across Codex, Claude Code, direct-model, and other registered Agent

sessions without losing Goal state or evidence;

• review protected changes through typed preview, explicit confirmation, and

receipts while LoopX state—not the browser—remains authoritative.

For Manager group conversations, LoopX keeps message visibility separate from Turn authority; see the bilingual Lark Manager context and authority contract.

loopx dashboard

loopx dashboard is the supported browser/PWA launch path. You can also download native desktop previews from the current release; they reuse the same loopback services and Goal state. Apple Silicon macOS supports signed App updates that pair the shell with its bundled runtime, plus repair and recovery. Python 3.11+ is required; the App is ad-hoc signed, not notarized. Windows NSIS preview installers support the same signed update flow on the main channel, with a separately installed CLI; the stable channel and MSI installs stay manual, App rollback and repair are unavailable, and the installer carries no Authenticode certificate (Windows may show a SmartScreen prompt). Desktop installation, updates, and source development.

Capability settings and reproducible workspace scenarios

From a source checkout, run python -m demo.workspace serve to explore a community event, a home-energy comparison, and a neighborhood website release. Each has four work roles, 18 tasks, two decisions, and two watches. The screenshot above comes from this reproducible workspace. Scenarios and replay instructions.

· Read the workspace guide · Try the five-minute tour

Try LoopX

Install once:

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

Start From Your Agent

Restart your Agent host, then start from a session opened on your project:

Host · Type Codex App, Codex CLI · $loopx Claude Code · /loopx DeepSeek Harness · Install the native DSH plugin, select the loopx skill, then describe the task

For example, give your Agent a concrete outcome:

$loopx speed up this project's slowest workflow; compare approaches and verify the gains

LoopX reuses existing project state or connects the project, plans the work as Todos and sets up the host's loop. loopx status shows the goal, any decision waiting for you and the next Todo; loopx doctor diagnoses the install.

Requires Python 3.11+ and Node.js 22.22.3+ (24 LTS recommended). Installing LoopX covers native Windows, pipx, upgrades, rollback and uninstall.

Upgrading from 1.x? Read the 2.0 migration and recovery guide before continuing existing Goals.

Other hosts, loop drivers and custom runners

Host · Recommended start · Loop driver Codex App · $loopx , or choose loopx from /skills, in a thread opened on the project. · Codex App heartbeat automation, refreshed from quota should-run.scheduler_hint Codex App over SSH · loopx agent-onboard --agent-type codex-app-ssh --project . · The returned visible /goal Codex CLI · Start codex in the project, then $loopx or /skills. · Visible /goal ; no hidden headless execution by default Claude Code · Install the opt-in adapter, then run /loopx followed by /loop. · Native Claude Code /loop gated by LoopX KunlunCode · Run loopx-kunluncode connect --project . --goal-id --agent-id , add a bounded todo, then run loopx-kunluncode run --project .. · Native Goal Pro through app-server; LoopX writes completion and quota only after strict verification OpenCode · Install the static command facade; opt in to --with-goal-bridge for recurring goals. · OpenCode command facade and explicit goal bridge Pi · Install the opt-in goal extension with loopx slash-commands --install --surface pi, then use /loopx from a trusted Pi session. · Visible Pi goal extension gated by LoopX quota (loopx_goal_activate + agent_settled continuation) ZCode · Install the skill facade with loopx slash-commands --install --surface zcode, then invoke the $loopx skill (or /loopx ) from a ZCode session in the project. · The ZCode session's own turn loop; every continuation enters through quota should-run Antigravity CLI (agy) · Install the skill facade with loopx slash-commands --install --surface agy, then invoke the loopx skill (or /loopx ) from an agy session in the project. · The session's native /goal loop (audited until `) with schedule self-wakes while the session lives; the facade instructs every turn/wake to re-enter through quota should-run — advisory pacing, not a host-enforced gate Kiro CLI · Install the skill facade with loopx slash-commands --install --surface kiro-cli, then run /loopx from a kiro-cli session in the project. · The session's native /goal --max Done when: loop, with the acceptance criteria stated inside the goal statement because the host derives them from it, bounded by the host's own iteration budget (default 5) and settled through the built-in goal completion contract; the facade instructs every turn and iteration to re-enter through quota should-run — advisory pacing, not a host-enforced gate DeepSeek Harness (dsh) · Install the native DSH plugin, select the loopx skill, and describe the task. The dsh goal-mode adapter remains available for headless turns. · Native same-session continuation and GoalBar, or headless dsh segments; both remain gated by LoopX authority Cursor, shell, or custom runner · loopx start-goal --guided --project . --goal-text ""`, or call LoopX from your runner. · Your shell, scheduler, or runner

The exact, copy-ready setup messages and host recovery paths live in Getting Started. Host integrations can inspect the Codex App host command registry contract, the Codex CLI packaged install path, the Claude Code adapter, the KunlunCode native Goal adapter, the Kiro CLI goal-mode adapter, or the DeepSeek Harness turn adapter.

See the 60-second DSH × LoopX Replan recording and reproducible fixture for the native path: install the plugin, select one skill, change a material constraint, and inspect the preserved decision trail.

For custom runners, start with the (python3 examples/custom-runtime-minimal-cli-turn-smoke.py), then the full guide and the worker bridge install contract. The core tick is deliberately small:

loopx quota should-run      # should this registered agent act now?
loopx todo claim            # who owns this slice?
loopx todo update           # what changed?
loopx refresh-state         # what should the next turn see?
loopx quota spend-slot      # account for a completed, validated slice

Clone-based install is only for contributors who want the live canary wrapper:

git clone https://github.com/loopx-project/loopx ~/loopx
~/loopx/scripts/install-local.sh
loopx doctor

First-Run Feedback

If LoopX works for you, a one-minute public issue helps us learn what a real first run looks like. It is optional, contains no telemetry, and should not include logs, paths, credentials, internal project names, or goal contents:

• First-run feedback
• Usage story for longer runs

loopx first-run-report prints the same prefilled link locally without sending anything.

Basic usage statistics default on after first-use disclosure: a daily random-ID heartbeat for platform support and continued use, plus separate ID-free CLI counts. No content is collected. Disable both in Settings → Capability Center or with loopx usage-ping disable / LOOPX_USAGE_PING=0; inspect payloads with loopx usage-ping status. See Basic usage statistics.

Evidence

LHTB: Higher Mean Reward With the Same Model

LoopX 1.0.3 Heartbeat reaches 0.4948 mean reward across 46 matched tasks: +17.3% over Plain Codex and +10.6% over native Codex Goal. All three use GPT-5.6 Sol, max reasoning effort, and disabled Web Search. Long-Horizon Terminal-Bench goes beyond coding: these tasks span research reproduction, scientific simulation, multimodal analysis, professional workflows, games, and systems work.