BloomGauge vs. darkbloom-manager
Updated
We make BloomGauge, so weigh this accordingly. darkbloom-manager is a Python command-line tool by benbuschmann on GitHub that keeps one model warm and switches when public network data favors another. Everything we say about it comes from its README as of September 29, 2026. BloomGauge is independent and not affiliated with Darkbloom.
At a glance
| BloomGauge | darkbloom-manager | |
|---|---|---|
| Runs as | Mac app with a local dashboard (macOS 14+, Apple Silicon) | Python script in Terminal (Python 3.10–3.14, standard library only) |
| Decides by | What Macs with your chip and memory confirmed earning per model, and your own Mac’s history | Public pressure (requests in flight per warm Mac) × blended list price × a per-model weight |
| Switch rule | At least five similar Macs clearly earned more for two hours; minimum run time, at most 3 moves a day by default, memory, power, temperature and idle checks | At least 25% better after switch cost, 3 passing checks a minute apart, 45 minutes minimum warm time, provider idle |
| Default | Observe-only until you turn the Manager on | Prints decisions until you add --apply |
| How it switches | Through Darkbloom’s CLI, waiting out the drain | darkbloom start with one model (restarts the provider) and sets the preload list |
| Recovery | Restores the model after a failed load, restarts a drained provider | Optional: after 3 checks where the Mac answers locally but production says model_not_loaded, stops for 15 minutes and starts the best model |
| File cache purge | Before each switch, with one-time permission | Optional, with a sudoers line you add |
| Earnings tracking | Confirmed credits per model and hour, base rewards separate | Not a goal: its README says the score does not measure payouts |
| Network view | Demand, capacity and pricing in the app | Public scoring graphs site |
| Phone and several Macs | Phone in a browser over your own Tailscale; up to ten Macs in one view | One copy of the script per Mac |
| License | MIT | MIT |
| Latest release | Signed, notarized builds from bloomgauge.io | v0.1.17, September 28, 2026 |
The main difference: list prices or confirmed pay
darkbloom-manager scores each model from public data: how many requests are in flight per warm Mac, times a blended price (85% input, 15% output), times a weight you can change. It is simple, transparent and needs no account data. Its README is clear that the score doesn’t measure your Mac’s speed, real token mix or payouts. BloomGauge instead decides from confirmed earnings, both on Macs with your chip and memory and on your own Mac. That tracks what routing actually sends a Mac like yours, but it needs history and holds still until the evidence is strong.
Choose darkbloom-manager if…
- You are comfortable in Terminal and want a single script you can read, test and run under your own supervisor.
- You want every threshold under your control: check interval, averaging window, improvement percentage, switch cost, minimum warm time and per-model weights.
- You want its routing recovery, which takes a Mac offline for 15 minutes when production keeps saying the model isn’t loaded.
- You run Macs headless or want no app at all.
Choose BloomGauge if…
- You want decisions based on what similar Macs were actually paid, not on list price and request pressure.
- You want earnings tracking in the same place: confirmed credits per model and hour, base rewards separate, and a live Pulse meter against your usual rate.
- You want a dashboard, phone access through your own Tailscale, and up to ten Macs in one view.
- You want recovery from failed loads and drained providers without extra setup.
Run only one tool that switches models
Two managers will undo each other’s switches, and every switch usually costs a 5-minute base-reward period. If you want to try darkbloom-manager, leave BloomGauge’s Manager off (observe-only is the default) and use BloomGauge only to track what each model earns. That is also a fair way to compare the two.
Sources
- darkbloom-manager README and releases: github.com/benbuschmann/darkbloom-manager
- BloomGauge README: github.com/cookder/bloomgauge
How BloomGauge helps
BloomGauge is free and open source (MIT). Its Manager starts observe-only and shows why no switch qualifies yet, so you can watch its reasoning before you let it act.
Questions
What is darkbloom-manager?
A free, MIT-licensed Python command-line tool that keeps one Darkbloom model warm and switches when public request pressure times price favors another model by at least 25% for three checks in a row, after a minimum warm time of 45 minutes. It can also send hourly test requests and recover from routing stalls.
Should I use BloomGauge or darkbloom-manager?
darkbloom-manager suits people who want a transparent, configurable script in Terminal. BloomGauge suits people who want a Mac app that decides from confirmed earnings, tracks earnings per model and hour, and works from a phone. Use one of them to switch models, not both.
Can I use darkbloom-manager with BloomGauge?
Yes, if BloomGauge’s Manager stays off. BloomGauge then only tracks earnings and network demand, and darkbloom-manager is the only tool switching models.
Related
- Darkbloom apps for Mac providers compared
- BloomGauge vs. Darkbloom Monitor
- Which Darkbloom model should I run on my Mac?
Updated 2026-09-29. Still stuck? Ask in #bloomgauge on the Darkbloom Slack or contact us. BloomGauge is independent and not affiliated with Darkbloom.