The control plane for long-running AI agents

Long-running agents.Lasting progress.

Turn complex work into durable goals, coordinated tasks, and reviewable results. LoopX keeps your agents moving across sessions. You keep the judgment.

Download for MacSee in action

Apple Silicon preview (.app.zip) · Python 3.11+ required. Install and first launch

CodexClaude CodeDeepSeek Harness+ more
ONE GOAL · ACROSS SESSIONS LIVE
ACTIVE GOALShip the next release
On track
P0 · Agent workResearch and implementation
Running
GATE · Human gateMerge needs your approval
Waiting
EvidenceValidation and handoff written
Complete
Next safe actionKeep independent tasks moving while review waits

A workflow illustration. Your runtime executes; LoopX keeps the state.

Explicit authority

Agents move fast without silently taking control.

Durable state

The next session resumes from governed state, not memory.

Reviewable evidence

Every transition leaves a result another person can verify.

One governed lifecycle

Long-running work stays legible.

The runtime executes. LoopX preserves the control state that lets the work continue safely across sessions, hosts, and providers.

01

Lifetime goal

One durable direction

02

Authority gate

A concrete decision

03

Agent lanes

Scoped continuation

04

Evidence

A validated outcome

05

Next run

Recoverable progress

Evidence from real loops

200+ hours, still legible.

One public contribution sequence and one redacted owner-run showcase preserve delivery, decisions, evidence branches, and recovery across many bounded agent turns.

~/loopx/evidence — curated replay

Public-safe evidence
LoopX AgentPublic trajectory 01 · Open-source issue fix
$ loopx status --goal-id issue-fix
GOAL

Ship the focused fix and preserve reusable knowledge.

ACTIVE
P0

Focused PR delivery · bounded diff and validation

RUNNING
GATE

Reviewer authority · approve the focused change

WAITING
P1

Repository knowledge and terminal resume remain available

READY
EVIDENCE

Merge/open result linked to the next reusable discovery

WRITTEN
NEXT

Review the focused diff, then continue the capability lane.

SAFE
200+ hours elapsed lifetimeCurated replay from public evidence

Elapsed lifetime is wall-clock project time. The Auto ML material is a redacted owner-run showcase, not a production result, company or employer endorsement, or independently reproducible evidence.

Browse all cases

Explore LoopX

See the workspace. Explore the evidence.

Try the product guide, read the research, and browse the full case collection.

Product demo

Personal Workspace

A shared view of Goals, Tasks, Chat, and outputs. Watch the demo and start your own local workspace.

Watch demo & read guide Research & evaluation

SWE-Marathon

Compare three retained execution modes across 15 matched tasks, with results, costs, and study limitations.

Read the study Research & evaluation

LHTB × LoopX

Compare five execution mechanisms across 46 long-horizon terminal tasks and inspect where recoverable state helps.

Read the study Research · Chinese

DeepSWE behavior discoveries

Explore how domain hints affect implementation and validation in individual cases.

Read the behavior analysis Research · Chinese

DeepSWE × Sol

Read the historical 113-task results, continuation mechanisms, and patch-delivery checks.

Read the research brief Cases

All LoopX showcases

Browse public cases, interactive walkthroughs, and their evidence boundaries.

Browse all cases
Projection developer tools

Across providers, runtimes, and hosts

One governed loop across agent runtimes.

Codex, Claude Code, OpenCode, TraeX, Pi, and custom runners can share one stateful goal lifecycle while each agent stays inside an explicit scope.

01

Durable state

Goals, decisions, todos, claims, quota, run history, and handoffs survive session boundaries.

02

Cross-runtime

Change the executor or host without reconstructing the objective from chat history.

03

Gate aware

A blocked P0 stays visible while independent P1 and P2 lanes continue safely.

04

Outcome driven

Every run writes validation, ownership, review, and handoff evidence into the next loop.

Developer book

Choose your path into LoopX.

A bilingual, protocol-first guide for developers who want to understand the control plane, connect a real project, or contribute along the right ownership boundary.

Open the Developer Book
01

Understand the control plane

Build the model from sessions and Goals through durable state, authority, governed turns, and recovery.

Start with foundations 02

Connect an existing project

Let your agent establish the Goal, identity, Host, and Git boundary, then verify the result.

Follow project onboarding 03

Contribute to LoopX

Trace protocols to source ownership, choose the right contribution surface, validate it, and prepare a focused PR.

Open the contribution map

Manual fallback

Prefer the shell? Install LoopX manually.

Use this path when your current agent cannot run shell commands. The same official installer registers lightweight command entries for supported hosts.

terminal
$ curl -fsSL https://loopx-project.github.io/loopx/install.sh | bash
$ export PATH="$HOME/.local/bin:$PATH"
$ loopx doctor
$ cd /path/to/your-project
$ loopx connect
$ loopx status

✓ CLI health checked
✓ project state readable
→ finish host activation from the onboarding packet
Read the Quick Start Inspect installer