The bot watches the whole farm and tunes every card itself
Especially useful on rigs with many different cards at once: 3070, 3090, 4070 side by side. The bot searches core, memory and power limit separately for each one, verifies the result for two hours, and keeps watching afterwards — if a card drops, it diagnoses the cause itself, reboots the rig through an escalating sequence of actions, and restores a safe preset without you lifting a finger. Reports arrive in Telegram.
$ /status GPU 0 RTX 3070 core +225 mem -100 PL 106 67.8 MH/s · 104.8 W · 0.650 /W GPU 1 RTX 3090 core +150 mem 0 PL 171 113.4 MH/s · 171 W GPU 4 verifying candidate 1h 12m of 2h PSU 1200 W · headroom OK preset verified (120 min)
Live preset catalog
Open full catalog →Features
Everything a rig needs, in one bot
Plain standard-library Python, no extra dependencies. Runs as a service on the rig itself.
Per-card tuning
Every card is tuned separately; settings are stored by the card’s own ID and survive a reinstall.
2-hour verification
A preset is accepted only after running stably. A card with repeated drops gets a longer check.
Power-supply budget
Total power limit stays within the PSU’s capacity. In hashrate mode, power is redistributed between cards.
Recovery from drops
Parses Xid errors and driver hangs, sets an overclock boundary, and reboots the rig through an escalating sequence of actions.
Controlled from Telegram
Status, modes, limits, pause, update and history, all by command. Only your ID can use it.
Preset catalog
The best settings per card model as a starting point for new rigs. No UUIDs or rig names.
Mixed cards in one rig
One setting for every card doesn’t work
A typical case: 3070, 3080 Ti, 3090 and 4070 in one rig. Each has its own core and memory headroom, its own memory chips, its own efficiency. Tuning such a rig by hand takes weeks, and one shared profile either loses hashrate or crashes the weaker cards.
Every card on its own
Its own core, mem and power limit for each card. Settings are tied to the card’s ID, not the slot number, so they don’t get mixed up when cards are moved.
Power is split smartly
In hashrate mode the bot raises one card’s power limit at the expense of the others, finds each one’s maximum, then picks the best distribution across the whole rig within the PSU’s budget.
A weak card doesn’t drag the rig down
A drop is recorded as a boundary for the exact card that caused it. The rest keep their presets, and the bot leaves that one card some headroom.
How it works
From install to a finished preset
The first search takes time; after that the bot just keeps watch.
Install
Unpack the archive, run the setup wizard, enter the bot token and your ID.
Search
The bot varies core, mem and power limit, measures hashrate and power draw, and marks boundaries.
Verify
The best points run under load for two hours.
Run
The preset applies itself; a safe profile kicks in on failure.
Modes
Choose what matters most
The mode switches on the fly with the /mode command.
efficiencyEfficiency
Maximum hashrate per watt. Usually at a reduced power limit, the cheapest on electricity.
compromiseCompromise
Maximum hashrate while staying within 3% of the best efficiency. The default mode.
hashrateHashrate
Maximum hashrate within the PSU budget. Extra power limit is then trimmed back as long as the loss stays under 0.5%.
Start
Install in five minutes
Detailed steps with ready-to-paste commands, a Telegram setup wizard, and a one-command update.
Open the guide# on the rig, as root cd /root && unzip -qo hive-oc-bot.zip cd hive-oc-bot && bash install.sh # update python3 -m ocbot --config config.json --self-update
FAQ
The essentials, briefly
Which cards and systems are supported?
Right now, NVIDIA cards on HiveOS; AMD support is coming. The algorithm and miner are set in the config.
Do I need an external server?
No. The bot runs on the rig itself, data is stored in a local SQLite database. Telegram is only needed for control and reports.
What if the rig drops during the search?
The bot records that point as a boundary, restores a safe profile, and continues with some headroom from it.
Is it safe to update?
Yes. Configs and the database are untouched; the archive is verified before the code is replaced, and a rollback is available if something goes wrong.