01
Watch every surface
One timeline for Codex, Claude Code, and CI—whether they run in a terminal, app, or hosted workflow.
The night shift for your AI agents
Keep agents moving.Never outside the lines.
Sleep Mode monitors Codex, Claude Code, and CI while you step away—recovering routine stalls and guarding every permission boundary.
Built forOpenAI CodexClaude CodeGitHub Actions
Goal progress5 of 8 checkpoints verified
All systems are inside policy. One approval is waiting for morning.
CX
CodexWorking
Implementing signed event ingestion
CL
Claude CodeHeld safely
Security review of device pairing
GH
GitHub ActionsWorking
Running unit and integration checks
Routine stall recoveredCodex resumed after a stale test fixture was corrected
8-hour durable runs Immutable permission ceiling Remote approvals Complete audit trail
Supervision, not wishful automationYour agents keep their tools.
Your agents keep their tools.
Sleep Mode keeps the watch.
A small local companion understands each agent's native events. The cloud control plane holds the goal, policy, timeline, and recovery loop.
02
Recover routine stalls
Detect failed checks, idle loops, and unanswered prompts. Retry or nudge only when policy says it is safe.
03
Hold the hard line
Permissions are concrete, signed, and rechecked on your device. Sleep Mode can ask for more—not grant itself more.
Your morning brief
Wake up to evidence, not a wall of terminal output.
Verified work, decisions, spend, risks, and the exact place a human needs to take over.
Good morning, WilliamNight shift made solid progress.
5 / 8 Event ingestion implemented and tested
Device pairing threat model reviewed
Draft PR held for your approval
Built for a mixed agent stackOne run can cross the CLI,
One run can cross the CLI,
editor, and CI.
Each connector advertises only what its provider can support safely. Observe and Nudge are broad; Operate must be earned action by action.
CX
Compare every integration OpenAI CodexNative · Codex App Server
CLClaude CodeNative · Structured process protocol
CUCursorNative · Structured stream JSON
GEGemini CLIACP · Agent Client Protocol v1
OCOpenCodeACP · Agent Client Protocol v1
COGitHub Copilot CLIACP · Agent Client Protocol v1
GHGitHub ActionsCloud · Signed GitHub webhooks
An operating model for agent work
Not just “keep the terminal alive.”
Sleep Mode separates monitoring, recovery, approval, and incident response so an unattended run can stop safely as well as continue safely.
01 · Overnight execution
Run AI coding agents overnight without granting a blank check.
A safety-first overnight supervisor for Codex, Claude Code, Cursor, Gemini CLI, OpenCode, Copilot CLI, and CI.
Explore use case 02 · Unified monitoringMonitor every coding agent from one durable control plane.
Unify local CLI agents, desktop coding tools, and GitHub Actions into a normalized, evidence-backed operations timeline.
Explore use case 03 · Remote approvalsApprove one AI agent action—not an entire unattended session.
Review exact coding-agent actions remotely with signed policy context, bounded expiry, and crash-safe one-time consumption.
Explore use case 04 · Failure recoveryTurn silent agent failures into bounded incident response.
Detect disconnected supervisors, repeated failed interventions, CI regressions, and exhausted budgets before they become runaway overnight jobs.
Explore use caseQuestions before the night shift
Know exactly what stays awake.
01Does Sleep Mode control my entire computer?
No. A per-user Rust daemon supervises explicitly armed provider processes and executes only typed, signed actions inside the selected workspace and local policy ceiling.
02Does it bypass Codex, Claude, Cursor, or other agent permissions?
No. Sleep Mode works through native structured protocols and keeps each provider's sandbox intact. Unsupported or ambiguous permission requests fail closed.
03What happens if the computer sleeps or loses its network connection?
The daemon holds an OS-native idle-sleep assertion while a run is armed and journals encrypted events locally. The cloud watchdog pauses the run after four minutes without a newly received signed event.
04Can an agent increase its own autonomy level?
Never. The run's signed permission ceiling is immutable. An agent may request one action already inside that ceiling, but only a human can arm a broader future policy.
Put your agents in Sleep Mode.
Start with observation. Earn trust one permission at a time.