NewGPU agents and pay-per-minute compute The operating system
The operating system
for your company's AI
Run your AI agents the way your company runs: departments with roles, one workspace shared with your people, and receipts for every action. On dedicated infrastructure that scales with every agent, and you pay only for the minutes they run.
Dedicated instance, never sharedFrom €49/month + usageAgents from €0.009/hour
RunsClaude CodeCodexGemini CLI
brainSYSTEM
acme/brain
Running
▶ consolidating graph12,408 nodes
Activity
312
Context
51%
Success
97%
High memory€38.10 · 2m ago
chief-of-staff
acme/chief-of-staff
Running
cron 07:00 ▶ morning brief
Activity
74
Context
46%
Success
92%
Standard€11.40 · 14m ago
One company · one operating system
Organised the way your company is.
ExoHuman draws your business as departments, each with its own agents and reporting lines. Every agent holds a role defined once, with objectives it is measured against and the people it answers to. Agents hand work to each other along the lines you draw, and nothing crosses a line you have not granted.
- Leadership · 2 agents
- Sales · 1 agent
- Marketing · 2 agents
- Customer support · 1 agent
- Operations · 2 agents
Leadership
brainSYSTEM
acme/brain
Running
▶ consolidating graph12,408 nodes
Activity
312
Context
51%
Success
97%
High memory€38.10 · 2m ago
Leadership
chief-of-staff
acme/chief-of-staff
Running
cron 07:00 ▶ morning brief
Activity
74
Context
46%
Success
92%
Standard€11.40 · 14m ago
Sales
sales
acme/sales
Waiting
⏸ approval · renewal offer31 drafted
Activity
129
Context
70%
Success
84%
Standard€16.60 · 3m ago
Marketing
marketing
acme/marketing
Running
▶ drafting campaign2 in queue
Activity
61
Context
44%
Success
88%
Standard€9.05 · 41m ago
Marketing
creative-studio
acme/creative-studio
Running
▶ rendering product shotsGPU · 14 min
Activity
38
Context
33%
Success
95%
GPU · L4€16.10 · just now
Customer support
customer-support
acme/customer-support
Running
▶ ticket #4312queue 3
Activity
210
Context
39%
Success
96%
Standard€20.80 · just now
Operations
back-office
acme/back-office
Running
▶ invoice run14 paid
Activity
96
Context
62%
Success
81%
Performance€10.95 · 22m ago
Operations
data-science
acme/data-science
Scaled to zero
suspended · idlebilling paused
Activity
22
Context
0%
Success
93%
GPU · A100€4.20 · 3h ago
Elastic by design
Every agent gets the machine it needs. Only while it works.
Each agent runs in its own isolated container, scheduled onto worker pools that grow when work arrives and scale to zero when it stops. Give your research agent a GPU, your coding agent more memory, and leave the rest on Standard. Billing stops the moment an agent goes idle; its memory stays in git.
- Compute profiles from 2 vCPU to H100 GPUs
- Custom profiles: vCPU, memory, GPU, disk
- Autoscaling worker pools per profile
- Per-minute billing, idle agents cost nothing
- Budgets and limits per agent and team
- EU and US regions
acme.exohuman.cc · worker poolseu-central
Standard 2 vCPU · 4 GB19 agents running · 6/12 nodes
Performance 4 vCPU · 8 GB4 agents running · 2/8 nodes
GPU · L4 1× L4 24 GB1 agent running · 1/4 nodes
GPU · A100 1× A100 80 GBscaled to zero
Running time this month1,284 h · €212.40 so far
Compounds
The work compounds.
What your agents learn every day becomes knowledge the company keeps: decisions, playbooks and facts, versioned and readable, without retraining a model. Every department feeds it; every agent reads from it. The longer it runs, the more of your company it holds, and that asset stays yours, exportable at any time.
Company knowledge
read by 23 agents · written by decision
- notes
- 12,408
- notes
- decisions
- 1,932
- decisions
- playbooks
- 214
- playbooks
Company facts, policies and positioning are written once, read by every agent, and changed only by decision.
With receipts
Every agent has a history you can read.
Each agent keeps its memory in a git repository: every decision, note and fix is a commit you can read, audit or roll back. Every action lands in an append-only audit log, with a hash chain you switch on in compliance mode.
Memory commits back to git on a schedule. Update is a push; undo is a revert. Replace the container and the mind survives in the repo.
isolated containerencrypted keyscpu / memory / gpu caps
acme/chief-of-staff · memory● committed 4m ago
- a3f19cmemory: 14 leads scored, 2 objections learned
- b7e02dmemory: weekly brief template refined
- c91a44skill: follow-up cadence v3
- d15f08memory: pricing objection playbook
Audit log · append-only · hash chain onchain verified
| #4824 | 09:13:40 | marketing | git.commit | auto |
| #4825 | 09:14:11 | back-office | crm.update | auto |
| #4826 | 09:14:56 | research | report.publish | approved · eugene |
| #4827 | 09:15:20 | brain | graph.merge | auto |
Shaped to your company
Anything irreversible waits for a person.
Money, client deliverables, deletions: an agent that reaches one of them stops and asks the person in the approving role, in their Inbox, before it acts. The answer wakes it back up. Objectives have an owner and a number, measured from what the agents record.
acme.exohuman.cc · Inboxmaria · Head of Sales
Action · 2UnreadAll
salesneeds approval
Northwind renewal: send the offer at 12% discount?
Finance recommends 12% (margin holds at 61%). Sales asked for 15%. The e-mail is drafted and goes out only when you answer.
email.send → [email protected] · frozen to this exact action
✓ Send at 12%Send at 10%Hold — I'll call them
Approved by maria · 09:12 · sales woke up and sent it · logged #4821
- QuestionWhich three accounts go in the board pack's risk section?12m
- Needs approvalPay invoice #2231 — €4,800 to Globex Ltd1h
- Came backCampaign brief v3 — deliverable ready2h
Where your people work
One workspace. Every agent.
Your team works with agents the way they work with colleagues: a chat per topic, a live card while the work happens, deliverables as files, a canvas with the numbers, ratings on what comes back. Also over Slack, WhatsApp, Telegram and voice.
acme.exohuman.cc · Workspace[email protected]
Pull the Q4 pipeline against target and draft the board pack. Loop in Sales for the renewal numbers.
On it. I asked sales for the renewal forecast and back-office for the cost lines; both came back in the last four minutes. The canvas has pipeline vs target; the pack draft follows.
Working · section 4 of 6, renewalsStandard · 2 vCPU
PDFQ4-board-pack-draft.pdf · 14 pagesrate it ▲ ▼
also overSlackWhatsAppTelegramVoiceMCPREST API
Hosted or hybrid
Our elastic cloud, plus your machines as worker nodes.
Both run the same ExoHuman, so you can move work between our capacity and yours without starting over.
Self-serve
ExoHuman Cloud
A dedicated instance we run for you on elastic infrastructure. Choose CPU or GPU profiles per agent, scale as you grow, and pay only for running time.
- Live in minutes after checkout
- Your own URL at <you>.exohuman.cc
- Backups, upgrades and monitoring included
- GPU agents on demand
Bring your own nodes
Your infrastructure, joined to the instance
Connect the machines you already own — any Linux server, VM or workstation — as worker nodes in two steps. Agents schedule onto them with quotas, approvals and the full audit trail.
- Two-step enrollment from the admin panel
- Your GPUs, storage and network stay yours
- Mix cloud profiles and your own capacity
- Same governance on every node
01
Create your account
Name, work e-mail and company. No sales call.
02
Configure the instance
Pick its address, region, plan and the compute profiles your agents need.
03
Pay securely
Checkout through Stripe. A small monthly base plus metered usage.
04
Open your URL
Your dedicated instance comes up; you are its first administrator.
Your fleet can be running before your next meeting.
Start your instance