You hit your session limit at 11pm. Instead of the session sitting dead until you wake up and type "continue," Claudify resumes it for you the second your quota resets — but only for sessions you explicitly opted in, only within the permissions you granted, and following a plan Claude wrote before you left.

No cc-tap, no browser automation, no undocumented APIs. Claudify drives the official claude -p --resume CLI and reads the transcript files Claude Code already writes to ~/.claude/projects/.

How it works

1. You opt a session in. In the dashboard, pick a session and set what it may do unattended (or, for a checklist-driven away plan, run /away inside a Claude Code session). Nothing runs without your explicit grant.
2. The daemon watches. It tails the session's transcript. When it sees You've hit your session limit · resets 6pm (Asia/Calcutta), it computes the reset time — spending zero quota to do so.
3. It resumes at reset. At the reset time it runs claude -p --resume <id> "<continuation prompt>" with your granted permission mode and turn caps. The resumed session reads .claudify/AWAY_PLAN.md, works the task queue, appends progress, and stops.
4. You get a notification and a morning-readable log of what happened.

Quick start

Everything below is done from the dashboard (a local web page) — no need to live in the terminal after install.

1. Install

npm install -g @bhavyabibra/claudify

Requires Node.js ≥ 20 and the Claude Code CLI (claude) installed.

2. Start the daemon (the background watcher)

claudify daemon install

On macOS this registers a launchd agent that runs at login, restarts on crash, and hosts the dashboard for you. On Windows/Linux, run claudify daemon run instead and keep it running (see Platform support).

3. Open the dashboard

claudify dashboard

This opens http://127.0.0.1:4177 in your browser. Bookmark it — it's your control panel. (If the daemon is already hosting it, this just opens the tab.)

4. Connect Claude Code

Click Connect Claude Code in the dashboard. A browser window opens; approve it. This signs Claudify in with your Claude subscription — a one-time step.

Claudify needs the standalone claude CLI's own login, which is separate from the Claude desktop app. The Connect button handles it; you don't need a terminal. (Under the hood it runs claude auth login.)

5. Arm a session

In Add a session, you'll see your recent Claude conversations — the same list as your Claude app sidebar. Click one and set:

• When to resume — when it hits the limit (the default) or at a set time (useful if you're near your limit but haven't hit it yet and know your reset time)
• How much it can do — plan / edit files (recommended) / run freely / all tools
• Spend limit, work per resume, times to continue, turn off after
• Optionally, a goal (what to work on) — treated as an objective, not a script, so repeated resumes won't redo finished work

Click Turn on auto-resume. That's it — walk away.

When your limit resets, Claudify resumes the session, and the continued conversation shows up right back in your Claude app. You'll get a notification with what ran.

Leave your machine on (and ideally on AC power) for overnight resumes — a sleeping laptop can't run the resume.

Everyday controls

Mostly you'll use the dashboard, but everything is scriptable too:

claudify list                 # what's armed and each state
claudify status               # daemon health + upcoming resumes
claudify logs -f              # follow activity live
claudify on <id>              # re-enable a session
claudify off <id>             # pause a session
claudify remove <id>          # stop watching entirely
claudify doctor               # preflight checks (--auth also tests login)

The dashboard also has a Resume now button on each armed session, to trigger a resume immediately (handy for testing or a manual nudge).

The permission grant

Because an unattended session acts with real permissions, enabling auto-resume requires an explicit grant. Modes, from safest:

• plan — read/plan only, no writes
• acceptEdits (recommended) — edit files; risky shell still blocks
• dontAsk — no prompts, limited to your allowlist
• bypassPermissions — full autonomy (requires --yes-i-accept-full-autonomy)

Plus caps: max turns per resume, max resumes per limit cycle, and a grant expiry. All enforced by the daemon. See SECURITY.md for the full model.

The away plan

/away writes .claudify/AWAY_PLAN.md in your project:

# Away Plan — myproject — 2026-07-14
## Objective        (definition of done for tonight)
## Task queue       (ordered checklist, worked top to bottom)
## Constraints      (what not to touch)
## Stop conditions  (when to stop early)
## Progress log     (each resumed run appends what it did / what's next)

Each resume continues from the top unchecked item and appends to the progress log, so the next run — or you, in the morning — picks up cleanly.

Requirements

• macOS, Windows, or Linux — see platform support
• Node.js ≥ 20
• Claude Code CLI (claude) installed (the dashboard's Connect button logs it in)
• A Claude plan with session limits (Pro/Max)

Platform support (honest status)

Claudify runs on macOS, Windows, and Linux — session detection, resumes, the desktop-app session list, the dashboard, and notifications are all platform-aware. The one thing that isn't fully automated everywhere is the always-on daemon:

| Piece | macOS | Windows / Linux | |---|---|---| | Session list, resumes, binary lookup, dashboard | ✅ | ✅ | | Notifications | osascript | PowerShell toast / notify-send | | Auto-start the daemon at login | ✅ claudify daemon install (launchd) | Run claudify daemon run, wired into Task Scheduler (Windows) or a systemd user service (Linux) |

So on Windows/Linux everything works; you just start the daemon yourself (one command, or wire it into your OS's startup). Native auto-start installers for those platforms are a welcome contribution.

Development

npm install
npm run typecheck
npm test
npm run build
npm run dev -- list        # run the CLI from source

Architecture: everything that touches Claude Code (CLI spawning, transcript parsing, limit-message wording) is isolated in src/driver/. The scheduler, store, and daemon know nothing Claude-specific — if Anthropic ships an official session API, only the driver changes.

Status

v0.1 — published and proven live. The full loop has been verified end to end on real hardware: limit detection, scheduled and on-limit triggers, real headless resumes that reappear in the Claude desktop app, spend caps, and idempotent continuation. Available on npm as @bhavyabibra/claudify.

Known limitation: auto-start installers for Windows/Linux aren't built yet (the daemon runs there via claudify daemon run). Contributions welcome.

License

MIT — see LICENSE.