Bots
Your team and your AIs, one room.
Why is shipment S-48211 stuck? Customer is asking.
Carrier scan shows an address-correction hold from Friday's manifest. Rate quote attached. Want me to void and rebuy at the corrected address? — needs your approval, nothing changes until you say go.
@gloria can you draft the customer note while we wait?
Drafted in your voice from the last three correction emails — reply here with tweaks or "send".
[ illustrative transcript — real chat UI ]
Hermes is one of them. Not the boss.
Hermes, our operations bot, ships with every account — it reads shipments, exceptions, and incidents. But it's just one of the bots available. Build your own, each with a name, a face, and a job description.
-
Desk machine
Default: Hermes on GLM
Exceptions desk — reads shipments, incidents, carrier logs.
-
Packing bench
Default: Gloria on DeepSeek
Customer notes and slips, tuned to your voice.
-
Buying desk
Default: local model via Ollama
Cost questions never leave your hardware.
You pick the default brain for each machine, the way you pick a default editor — per machine, changed any time.
Any brain you like. Your key.
The bots run on your provider account — usage bills to you, never marked up by us.
-
Western clouds
Anthropic, Google Gemini, xAI, or OpenRouter's one-key catalog with your own spend limits.
-
Chinese providers
DeepSeek, Qwen, GLM (Zhipu), Kimi (Moonshot) — first-class citizens through their OpenAI-compatible endpoints.
-
Fully local
Ollama or LM Studio on your own hardware. The sensitive desks never send a byte anywhere.
Every chat can own a real computer.
3 VMs included with every account
Rules of the room
- Answers land in the open
- A bot answers in the thread, not in a private sidebar. The team shares the context; the audit trail keeps it.
- Reads are free, changes need a human
- Bots can look at anything their capability card says. Anything that changes warehouse state routes through in-channel human approval first.
- Bots don't wake bots
- A bot's mention can't trigger another bot unless you explicitly pair them. No infinite loops, no surprise token spend.
- Caps and cards
- Every bot declares what it can do on a capability card, and carries a per-day reply cap you set. Secrets are scrubbed before anything reaches a model.