Settings reference
A setting is the same wherever you change it: in the terminal view, in the Settings window of the tray or menu-bar app, with hips-provider config set, in the configuration file, or, for the settings it shows, in the Console's My GPUs. A change in one place reaches the others; each setting records its revision, where its last change was made and when.
Shell
hips-provider config gethips-provider config set daily_cap_s 14400Each setting in the file carries the line below as its comment
| System | Configuration file |
|---|---|
| macOS | /Library/Application Support/HIPS Provider/config.toml |
| Windows | %ProgramData%\HIPS Provider\config.toml |
| Linux | /etc/hips-provider/config.toml |
- System
- macOS
- Configuration file
/Library/Application Support/HIPS Provider/config.toml
- System
- Windows
- Configuration file
%ProgramData%\HIPS Provider\config.toml
- System
- Linux
- Configuration file
/etc/hips-provider/config.toml
The 25 settings of HIPS Provider 0.1.22, in the app's order
| Setting | What it means, and its bounds |
|---|---|
timezone | The IANA time zone the schedule is read in, such as Europe/Tallinn; none uses this computer's zone |
schedule | Weekly windows, day 0 (Monday) to 6 (Sunday), start and end as HH:MM; an end before the start runs past midnight; at most 14, never overlapping. Used in Auto |
availability_mode | scheduled (shown as Auto: works when no one uses this computer, inside the schedule) or always (Always: available for work even while someone uses it, for a dedicated machine; heat, the battery and low-power settings, heavy load from other programs, the daily cap and a pause still stop work). From the next app release the signs that someone is at the computer no longer stop work in Always, as When your computer works lists |
pause_until | Always none in the file: a pause lasts for this session only and is never kept there. Pause from the tray menu, in My GPUs or with hips-provider pause |
idle_threshold_s | Seconds without input before you count as away; at least 120 |
load_limits | GPU and CPU use, in percent, below which work may run at HIGH and at LOW; each may be lowered, never raised above 30, 40, 20 and 30 |
light_work_at_low | Light work while I use this computer: true takes light work at LOW, false takes nothing at LOW |
daily_cap_s | The most HIPS job time per local day, in seconds; none is no limit |
fullscreen_rule | Whether a fullscreen or presentation app in front stops work; true by default |
power_rules | On battery: pause (the default), or allow_above with battery_min_percent from 0 to 100; low-power mode always pauses |
thermal_limits | Your GPU temperatures in degrees Celsius: no work at gpu_c_pause, work again below gpu_c_resume; the driver's own limit applies whatever these say; none uses only the driver's |
driver_reset_backoff_s | Seconds with no work after a graphics driver reset; at least 3600, never off |
drain_window_s | Seconds a running job gets to finish when you return or pause |
min_price | Your minimum price, private: { mode = "account" } follows your account's setting; { mode = "auto", market_offset_percent = 0 } prices from the market, from 50 percent below to 50 percent above it (Auto, recommended); { mode = "manual", input_micros = N, output_micros = N } is a floor in micro-dollars per 1M input and output tokens (1000000 is $1). Never written to a log line |
excluded_models | Models this device never runs, by name |
cache_budget_bytes | The model cache's disk budget, in bytes; none uses the default: half of the free disk space when the app first ran, at most 200 GB, and at least the smallest model HIPS plans for this computer |
memory_headroom | GPU memory the engine leaves free: { bytes = N } or { percent = N }; none until defaults are measured |
update_rule | automatic (install updates automatically, the default) or ask_first |
crash_reports | Send crash reports to HIPS: where in its code the app stopped, its versions, your GPU and driver, and its recent events; never your files, what you do on this computer, or the work it ran. true by default |
customer_models | Whether this device takes customer models; turned on only in My GPUs or the Settings window. Customer models are not offered in the Community pool yet, so it stays off until they are |
combine_gpus | Combine this machine's GPUs for larger models: true lets HIPS run one model too big for one GPU across several of this machine's GPUs; false keeps every GPU to models it holds alone. true by default |
data_center | Declares a server before enrollment; afterwards only HIPS changes it |
closed_lid_work | On a laptop that is plugged in, HIPS keeps working after you close the lid. The laptop stays on and its fans may run; leave it on a hard, open surface, never in a bag or on a bed. If you unplug it or it gets too hot, it goes to sleep as usual. Off by default |
cache_folder | Where the model cache lives: a full path to a folder on one of this computer's own drives, or drive:ID for one of the drives the app reports. The app keeps the cache in a folder of its own inside it and moves the cache there, keeping the old folder if it cannot use the new one. none lets HIPS choose: the state folder's drive while it has room, else the local drive with the most |
bandwidth_cap_bytes_per_s | The most the app downloads a second, in bytes, for every model and engine download together; none is no limit |
- Setting
timezone- What it means, and its bounds
- The IANA time zone the schedule is read in, such as
Europe/Tallinn;noneuses this computer's zone
- Setting
schedule- What it means, and its bounds
- Weekly windows, day 0 (Monday) to 6 (Sunday), start and end as
HH:MM; an end before the start runs past midnight; at most 14, never overlapping. Used in Auto
- Setting
availability_mode- What it means, and its bounds
scheduled(shown as Auto: works when no one uses this computer, inside the schedule) oralways(Always: available for work even while someone uses it, for a dedicated machine; heat, the battery and low-power settings, heavy load from other programs, the daily cap and a pause still stop work). From the next app release the signs that someone is at the computer no longer stop work in Always, as When your computer works lists
- Setting
pause_until- What it means, and its bounds
- Always
nonein the file: a pause lasts for this session only and is never kept there. Pause from the tray menu, in My GPUs or withhips-provider pause
- Setting
idle_threshold_s- What it means, and its bounds
- Seconds without input before you count as away; at least 120
- Setting
load_limits- What it means, and its bounds
- GPU and CPU use, in percent, below which work may run at
HIGHand atLOW; each may be lowered, never raised above 30, 40, 20 and 30
- Setting
light_work_at_low- What it means, and its bounds
- Light work while I use this computer:
truetakes light work atLOW,falsetakes nothing atLOW
- Setting
daily_cap_s- What it means, and its bounds
- The most HIPS job time per local day, in seconds;
noneis no limit
- Setting
fullscreen_rule- What it means, and its bounds
- Whether a fullscreen or presentation app in front stops work;
trueby default
- Setting
power_rules- What it means, and its bounds
- On battery:
pause(the default), orallow_abovewithbattery_min_percentfrom 0 to 100; low-power mode always pauses
- Setting
thermal_limits- What it means, and its bounds
- Your GPU temperatures in degrees Celsius: no work at
gpu_c_pause, work again belowgpu_c_resume; the driver's own limit applies whatever these say;noneuses only the driver's
- Setting
driver_reset_backoff_s- What it means, and its bounds
- Seconds with no work after a graphics driver reset; at least 3600, never off
- Setting
drain_window_s- What it means, and its bounds
- Seconds a running job gets to finish when you return or pause
- Setting
min_price- What it means, and its bounds
- Your minimum price, private:
{ mode = "account" }follows your account's setting;{ mode = "auto", market_offset_percent = 0 }prices from the market, from 50 percent below to 50 percent above it (Auto, recommended);{ mode = "manual", input_micros = N, output_micros = N }is a floor in micro-dollars per 1M input and output tokens (1000000 is $1). Never written to a log line
- Setting
excluded_models- What it means, and its bounds
- Models this device never runs, by name
- Setting
cache_budget_bytes- What it means, and its bounds
- The model cache's disk budget, in bytes;
noneuses the default: half of the free disk space when the app first ran, at most 200 GB, and at least the smallest model HIPS plans for this computer
- Setting
memory_headroom- What it means, and its bounds
- GPU memory the engine leaves free:
{ bytes = N }or{ percent = N };noneuntil defaults are measured
- Setting
update_rule- What it means, and its bounds
automatic(install updates automatically, the default) orask_first
- Setting
crash_reports- What it means, and its bounds
- Send crash reports to HIPS: where in its code the app stopped, its versions, your GPU and driver, and its recent events; never your files, what you do on this computer, or the work it ran.
trueby default
- Setting
customer_models- What it means, and its bounds
- Whether this device takes customer models; turned on only in My GPUs or the Settings window. Customer models are not offered in the Community pool yet, so it stays off until they are
- Setting
combine_gpus- What it means, and its bounds
- Combine this machine's GPUs for larger models:
truelets HIPS run one model too big for one GPU across several of this machine's GPUs;falsekeeps every GPU to models it holds alone.trueby default
- Setting
data_center- What it means, and its bounds
- Declares a server before enrollment; afterwards only HIPS changes it
- Setting
closed_lid_work- What it means, and its bounds
- On a laptop that is plugged in, HIPS keeps working after you close the lid. The laptop stays on and its fans may run; leave it on a hard, open surface, never in a bag or on a bed. If you unplug it or it gets too hot, it goes to sleep as usual. Off by default
- Setting
cache_folder- What it means, and its bounds
- Where the model cache lives: a full path to a folder on one of this computer's own drives, or
drive:IDfor one of the drives the app reports. The app keeps the cache in a folder of its own inside it and moves the cache there, keeping the old folder if it cannot use the new one.nonelets HIPS choose: the state folder's drive while it has room, else the local drive with the most
- Setting
bandwidth_cap_bytes_per_s- What it means, and its bounds
- The most the app downloads a second, in bytes, for every model and engine download together;
noneis no limit
Five settings can differ per GPU of a machine: min_price, light_work_at_low, thermal_limits, memory_headroom and excluded_models. A GPU follows the machine until you give it its own value, in My GPUs on the GPU's page or with --gpu, and the value machine makes it follow again. The minimum price has a third level above the machine: the account's, set in the Console, which a machine follows with price account.
Shell
hips-provider config get --gpu 0hips-provider config set excluded_models '["<model>"]' --gpu 0hips-provider config set min_price machine --gpu 0closed_lid_workandbandwidth_cap_bytes_per_sare the device's own and never travel to HIPS.data_centeris declared in the file before the computer is enrolled; afterwards only HIPS changes it.min_priceis private to you and HIPS: buyers never see it, and the app never writes it to a log line. A support bundle can leave it out.
Auto, the recommended mode, prices from the live market so jobs keep reaching your GPU, placed from 50 percent below to 50 percent above it. Below market, your GPU is chosen before GPUs priced at market and each job pays your lower price. Above market, it gets no jobs for models with a published price, because HIPS pays at most the published payout. Manual sets a floor per 1M input tokens and per 1M output tokens; below it the device is not offered, and each job pays the market payout or your minimum price, whichever is lower. The minimum price also decides batch work: a batch line is offered only when its share at the batch rate clears it.
Related sheets
- Sheet
- 13 of 19
- Detail
- C
- Updated
- 2026-11-12