No black-box bills.
Watch every agent's cost as it runs. Per-conversation,
per-turn, per-minute. Pause, kill, or dial back in one click. Hosted from
$9 a month, first one free.
💸 Cost per conversation
Every turn logged, drilled down to the token. Open
the Billing tab to see which conversation actually cost you $3.
⚡ Live spend, per agent
The Fleet view shows what each agent has burnt
today, updated as messages land. No waiting for the invoice.
🛑 Pause or kill, one click
Runaway agent? Stop it before the next turn.
Delete an agent and the unused month is refunded to credits.
🎁 Free to start: explore and design your agent in
chat with no account. Signing up adds a
$5 credit with no card required,
and your first agent's first month is free on Standard and Plus.
Standard
$9 /agent/mo
1 GB micro-VM: one-shot coding agents + picoclaw
- 1 GB isolated micro-VM per agent
- Fits one-shot coding agents and light chat bots
- Included model on credits, or bring your own API key
Plus
$19 /agent/mo
2 GB micro-VM: openclaw, nanoclaw, or any Standard agent upsized
- Everything in Standard, on a 2 GB micro-VM
- Runs OpenClaw and NanoClaw, the full-featured runtimes
- Headroom for sub-agents, crons, and heavier jobs
Dedicated boxes
$59+ /mo
NemoClaw-sandboxed multi-agent VMs for teams and compliance
- Secure Box $59: shared 4-vCPU/8 GB VM (+swap): 1-2 NemoClaw-sandboxed agents - NVIDIA's documented minimum
- Multi-Agent Box $199: dedicated 4-vCPU/8 GB VM: up to 5 NemoClaw-sandboxed agents
- Multi-Agent Box XL $299: dedicated 4-vCPU/16 GB VM: up to 10 NemoClaw-sandboxed agents
Every plan includes
Live observability of every token, tool call, and dollar
(the same stream as ClawMetry). Governance: exec approvals before dangerous
commands, per-agent budgets, pause and kill from the Fleet view. A browser
terminal with root on your VM. Encrypted secrets. Chat channels: Telegram,
Slack, Discord, web chat and more. And a self-host export kit, so the agent
is yours to take home.
The model meter, in plain terms
Every model token your agent spends is billed at the
provider's publicly listed price. We add nothing on top. Use the included
model billed from your credits, or paste your own Anthropic, OpenAI, or
Cursor API key and pay your provider directly while we host for a flat
monthly price. Either way the ledger traces every charge back to the agent
and the moment it happened.
Why flat monthly beats a compute meter. Platforms that bill compute
by the second price an always-reachable 4 GB agent at roughly $50 per month
before a single model token, and the meter anxiety is yours to carry. Here
an always-on agent is $9 or $19 flat, the model bill is the provider's own
rate, and the dashboard shows you both in real time. No surprise line
items.
Fair exit, in both directions
Retire an agent and the unused remainder of its prepaid
month comes back as credits, prorated to the minute. Want to leave hosting
entirely? Export the self-host kit and take the agent with you. It is your
agent.
Questions
- What happens when my credits run out?
- Your agent is stopped, never silently billed. Top up and it resumes
where it left off. Volumes and history persist for 30 days.
- Do you mark up model usage?
- No. Tokens are forwarded at provider cost. Hosting is the paid
product.
- Can I switch plans later?
- Yes, upsize or downsize anytime. The unused part of the old month is
refunded and the new plan starts fresh.
- Is my agent isolated?
- Each agent runs in its own micro-VM with hardware-level isolation and
sealed secrets. You get root on the VM; nobody else does.
ClawMetry Agent Builder