macOS 14+ · direct download
Know when Claude Code will hit its limit — before it stops your work.
Throttle puts your 5-hour and weekly usage in the macOS menu bar, predicts when you'll hit the cap, and keeps every reading local.
No account · No telemetry · macOS 14+ · Apple Silicon & Intel
No more limit surprises
A predictive "≈2h to your cap" nudge before the wall — never a lockout mid-task.
Many sessions, one calm Mac
Run every project from one cockpit; idle sessions hibernate before RAM runs out.
Cut invisible token waste
Bloated configs, verbose output, re-charged images — found, measured, trimmed.
Live usage meter
Your 5-hour & weekly limits, always one glance away.
Exact when you're signed in, a clearly-degraded estimate when you're not — never a confident number that might be wrong. A predictive nudge tells you ≈2h to your cap before the wall, not after.
What's inside
Five instruments, one calm panel.
Your 5-hour and weekly Claude limits in the menu bar — exact or estimated — with a predictive "you'll hit your cap in ~2h at this burn" nudge so a lockout never surprises you mid-task.
Run several claude projects at once — tabs, a project rail, or a mission-control overview — each showing real cost, model and RAM, under one shared binding limit and a memory-pressure gate that stops you oversaturating the Mac.
Keeps Claude Code lean system-wide — a concise output-style and a usage statusline in every terminal — so every session spends fewer tokens by default. Fully reversible, any time.
Proposes a leaner, safer CLAUDE.md as a GitHub-style diff with a plain "why this is better" summary. You review every line before anything changes.
Surgical context trimming, memory-pressure awareness and a config audit that flags what's quietly costing you — all measured, never guessed.
Fewer tokens
Optimizers that cut what you spend.
One-click reply styles — Concise, or aggressive "Caveman" compression — applied to every claude session, terminal and Cockpit. Less prose per answer, the code and reasoning untouched.
Deterministic one-click fixes — deny blind reads of .env, node_modules and build output, set a cost-sane default model — each a reviewable diff, backed up before any change.
Drop a screenshot into a Cockpit session and Throttle OCRs it locally to text — a fraction of the tokens of a vision image. Hold ⌥ to keep it as an image.
Strips re-charged base64 screenshots from past transcripts so --resume reloads light — reversible, the original backed up, the active session untouched.
Searchable session memoryNew
Search your past claude sessions for context the window has lost — a dedicated Search window, plus an opt-in MCP server so Claude can search its own history too. Finds the right session even when you don't recall the exact word — searching "auth" also surfaces "login" and "token", entirely on-device.
MCP manager & memory hygieneNew
Move any MCP server between Global, project-local and the shareable .mcp.json, enable, disable, add or remove one — each file backed up first. Zombie servers flagged with a real list_tools handshake; MEMORY.md indexed past Claude Code's 200-line / 25 KB auto-load cap.
Keeps your Claude Code setup optimized system-wide, by default — concise replies and a live usage statusline on, dead memory from deleted projects archived. Every action reversible and logged; review or undo any of it in one place.
A PostToolUse hook strips ANSI, dedups and trims verbose CLI output before Claude sees it — a green test run collapses to a line, errors always passed through raw. Fewer tokens per build, reversible.
Cache-bust attributionNew
Shows why your prompt cache is being re-written — model swaps vs prefix churn — so the silent 10× re-reads that quietly inflate a session stop being invisible. Computed on-device from your own transcripts.
Everything inside
One app. The whole cockpit.
Meter
Live 5-hour & weekly
Your Claude limits in the menu bar.
Predictive cap ETA
"~2h to your cap" before the wall.
Exact or honest estimate
Never a confident wrong number.
Reset countdown
Resets 9pm — in 2h 14m.
Cockpit — many sessions, one panel
Multi-session tabs
Rail + mission-control overview.
Memory-pressure gate
Stops you oversaturating the Mac.
Auto-hibernate idleNew
Reclaims RAM under pressure or when crowded.
Leak restartNew
Watches RSS, restarts in place, keeps context.
Predictive pacingNew
Warns before several sessions hit the cap.
State-aware pause
Freezes only at a quiet moment — reversible.
Duplicate consolidate
One click to merge stray sessions.
Rate-limit aware
Flags which session, counts down the reset.
Session offloadNew
Runs claude on your own box — frees the Mac's RAM.
Optimize — fewer tokens
Output styles
Concise or Caveman, every session.
Compress command output
Collapses noisy CLI output; errors raw.
Context trimmer
Strips re-charged images from past sessions.
Auto-trim on launchNew
Lighter --resume, reversible, opt-in.
Image → text on drop
Local OCR instead of a vision image.
Read Pressure → searchNew
Bulk file reads? One tap to local semantic index.
CLAUDE.md & settings hardening
Reviewable diffs, backed up first.
Cache-bust attribution
Shows why your prompt cache re-writes.
MCP control
MCP managerNew
Move/enable/disable/add across scopes.
On-device MCP advisorNew
Keep / disable / offload — 1-click Optimize.
Run MCP on your serverNew
Offload servers off the Mac; native verify.
Command runnerNew
Saved local commands, zero tokens.
Insight
Test outcomes / € per greenNew
Your working results, priced. Measure-only.
Searchable session memory
Find lost context by meaning, on-device.
Semantic project index
throttle_semantic_search, 100% local.
Autopilot
Keeps the setup lean system-wide, reversible.
It audits and optimizes — it never becomes the engine.
Throttle reads ~/.claude on your Mac and stays out of the way. 100% local — no telemetry, no analytics, no account: your prompts, code and usage never leave the machine. Signed & notarized, auto-updates via Sparkle; uninstall cleanly any time and every change reverts.
Pricing
Free to meter. Pro to optimize.
Free
€0
- Live menu-bar meter (estimated)
- Predictive cap nudge
- Single-session cockpit
- Exact server-true readings
- AI Optimizer & multi-session
Pro
€29 one-time
- Everything in Free
- Exact server-true usage
- Multi-session cockpit & gate
- Autopilot & AI Optimizer
- Stats history & projections
- Team policy export (MDM)
- Offload sessions to your own server
Never get cut off mid-task again.
Download the free meter and see exactly where your tokens go — in under a minute.
macOS 14+ · v3.2.71 · 13.4 MB · auto-updates via Sparkle