JURNITI°
Login
Live · provision in ~3 minutes · 30-day money-back

A fleet of agents.
Always on.

Your agent fleet in the cloud — each in its own secure microVM, your keys, always on. Your laptop stays closed. From $21/mo billed yearly.

Firecracker + KVM
Keys stay in your VM
Always on · laptop closed

See a working agenthow cohorts work

The problem

Most managed agents are not what they say they are.

The category default for managed AI agents is a Docker container with a friendlier dashboard. That stops being acceptable the moment your agent has the keys — this spring a single internet scan engine indexed more than 180,000 exposed agent instances.

Docker is not isolation.

Containers share one Linux kernel. Your neighbour's bug becomes your bug. The container escapes are documented and public — runc CVE-2024-21626 (“Leaky Vessels”) is the standing reminder.

Managed agents see your keys.

Most platforms proxy your model spend through their account. Lock-in by design + vendor markup on every token. You don't own the relationship with your provider.

Setup eats your weekend.

DIY agent infra means a box, firewall ports, TLS, persistent volumes, systemd units, log forwarding. Then you get to install the agent. A managed agent shouldn't ask you to be a sysadmin first.

Pick your harness

Autonomous agents. Coding agents. One platform.

19 open-source agent runtimes — from Claude Code and Codex to Hermes and OpenClaw — each in its own KVM-isolated microVM with your own keys. Pick one to start; add more later, each in its own VM, billed at its plan's rate.

Autonomous & always-on agents(5)

Coding agents(14)

Agentmaxxing

Deploy a fleet of 1k agents
on one command.

Choose public templates or private workflows. jurniti creates isolated workers, dispatches concurrently, and collects every result into ./out. Spot pricing makes a fleet pizza money, not payroll.

jurniti run swarm runtime fans out to Hermes, OpenClaw, Pi, and custom workers, then collects one file per agent into ./outJURNITI RUNSWARM RUNTIMEHERMESPUBLICOPENCLAWPRIVATEPIPUBLICCUSTOMPRIVATE./out/ONE FILE PER AGENT
Cohorts · why variants matter

Fifty agents. One SEO brief.

One agent fails confidently. Fan the same ranking brief across fifty isolated workers — titles, metas, internal links — then keep the fixes that hold up in ./out.

Up
50 isolated workers on Spot. Grouped as seo-week.
Run
Same SEO brief fanned out. Retries visible.
Cp
Every artifact lands locally. Rank and ship.
Example · SEO
# Create a cohort — ranking variants need isolation
$ jurniti up --tier spot --harness hermes --count 50 --group seo-week
✓ 50 microVMs ready

# Dispatch and collect — same SEO brief, many fixes
$ jurniti run --group seo-week --model $MODEL --provider openrouter "Stage title and meta fixes for my top 50 pages."
✓ 48succeeded  2 retrying
$ jurniti cp --group seo-week work/out.txt ./out
Features

Hardware isolation. Your keys. Your subdomain.

One Firecracker microVM per tenant — the same KVM hardware boundary AWS uses for Lambda, on nested-virt EC2 in us-west-2. Which means your keys, your disk, and your blast radius stay yours.

Real Firecracker microVM

KVM hardware boundary per tenant. The same primitive AWS uses for Lambda. 124ms boot.

Bring your own model key

OpenRouter, OpenAI, Anthropic, Nous Portal. Paste once, lives only inside your VM.

30-day money-back

Cancel within 30 days for a full refund. Manual processing within one business day.

Custom subdomain

you.jurniti.com with valid TLS, ready when the VM boots. Pro and Max plans.

Upstream Hermes

We ship uv and the upstream install script; you control which Hermes version lands.

Persistent volume

/var/lib/jurniti/persist survives restarts and re-provisions. Daily snapshot on Max.

Memory

Agents that remember — without you pasting the past back in.

Most agents wake up empty every session. jurniti Memory is a dedicated microVM that holds long-term knowledge, then plugs into any agent over MCP. The memory stays in your tenant. The agent just gets smarter.

Obsidian logo
Keyless

Obsidian

Markdown notes the agent can read and write. Portable, offline-friendly, zero model key to start.

Keyless

gbrain

A self-wiring knowledge graph. Ingest notes once; the agent gets a second brain that compounds.

mem0 logo
Semantic

mem0

Vector recall that actually finds what happened last week. Add your model key when you're ready for Pro memory.

Buy a memory once. Grant it to one agent or many. Sign in to provision — same Firecracker isolation as your agents, from $19/mo.

Observability

Stop grepping logs. Watch what the agent actually did.

When an agent fails mid-tool-call, the answer is not another SSH session. jurniti Observability is a self-hosted sidecar in its own microVM — Latitude or Langfuse — so sessions, tool calls, and multi-turn paths stay on your tenant. Same isolation model as agents and Memory.

Latitude logo
Self-hosted

Latitude

Agent-native traces, signals, and session trajectories. MIT, OTEL ingest, full UI in your microVM.

Langfuse logo
Self-hosted

Langfuse

The open-source default for LLM observability. Sessions, scores, and datasets — data residency you control.

Not a shared SaaS black box. A sidecar you own.

One observability engine. Many agents. Buy a sidecar, grant it from the agent VM page, and the platform injects OTEL export config — traces stay in your box. Cloud-only vendors stay off the menu on purpose.

After sign-in, open Observability in the dashboard to add an engine, then attach it from any agent — same pattern as Memory.

Quickstart

Three commands. Then your agent is online.

Run inside the microVM you provisioned. We don't pre-install Hermes — you control the version, the upstream installer is one curl away.

you.jurniti.com●
# Boot one tenant microVM
$ hermes config set OPENROUTER_API_KEY sk-or-v1-...
$ hermes dashboard --host 0.0.0.0 --port 9119

# 124ms later …
→ tenant.you.jurniti.com  ·  KVM-isolated
Command line

Install the CLI. Ship in one line.

One binary drives everything — provision, billing, VMs, templates — and doubles as an MCP server for your own agent.

$ curl -fsSL https://jurniti.com/install.sh | sh

Then, three commands to first agent

1. jurniti login

authenticate this machine

2. jurniti billing add-card

attach a payment method

3. jurniti up

provision + boot your agent VM

4. jurniti auth <vm>

add your model key, or sign in with --login

Read the docsWindows? irm https://jurniti.com/install.ps1 | iex
Pricing

Flat monthly. No metered surprises.

Every plan is the full platform — BYOK, no feature gates. Annual prepay gets 2 months free.

Starter

$21/moinstead of $25/mo at the monthly rate

$250 billed once a year — 2 months free

Solo builders running Pi or Hermes 8B with a local proxy. Just enough to keep your agent's home alive 24/7.

1 vCPU · 2 GiB RAM · 10 GiB SSD · 100 Mbit/s burst

  • Real KVM isolation — your own Firecracker microVM
  • BYOK — your keys never leave the box
  • Public TLS subdomain
  • Persistent home that survives restarts

Pro

Most popular
$41/moinstead of $49/mo at the monthly rate

$490 billed once a year — 2 months free

The default. Headroom for Hermes-3-8B at Q4, OpenClaw with a heavy plugin set, or a Pi instance you actually live in.

2 vCPU · 6 GiB RAM · 50 GiB SSD · 200 Mbit/s burst

  • Everything in Starter, plus
  • Custom subdomain
  • 2× the compute, 5× the disk of Starter

Max

$83/moinstead of $99/mo at the monthly rate

$990 billed once a year — 2 months free

For agents you'd cry about losing. Daily snapshots, room for a 12 GiB working set, and the headroom to keep multiple skills warm.

4 vCPU · 12 GiB RAM · 100 GiB SSD · 500 Mbit/s burst

  • Everything in Pro, plus
  • Daily snapshot
  • 2× the compute, 2× the disk of Pro
30-day money-back guaranteeBYOK — no token markupCancel anytimeLive in ~3 minutes

Don't need always-on? Pay by the hour with prepaid credits.

On-Demand

usage · prepaid credits

Never interrupted while running.

$0.05/hour

Metered per second while running · Starter sizing

Bursty and part-time agents. Run for an afternoon, stop, pay for exactly the hours you used.

  • Starts in seconds when warm capacity is available, otherwise queued — you'll be emailed when it starts.
  • Running 24/7 ≈ $36.50/mo — the $25/mo flat plan is cheaper for always-on.
  • Prepaid credits — $10 minimum top-up, card on file, never charged until you top up
  • Billed per second while running (60-second minimum per run) at Starter sizing (1 vCPU · 2 GiB RAM)
  • Stop anytime — a stopped agent burns nothing
Buy credits — from $10

Spot

usage · prepaid credits

Half the rate — can be interrupted.

$0.025/hour

Metered per second while running · Starter sizing

Interruption-tolerant work — batch runs, experiments, agents that checkpoint their own progress.

  • AWS reclaims can lose up to the last ~10 minutes of unsynced work; jurniti-initiated pauses lose nothing.
  • Pauses gracefully and restores automatically when capacity frees.
  • Prepaid credits — $10 minimum top-up, card on file, never charged until you top up
  • Billed per second while running (60-second minimum per run) at Starter sizing (1 vCPU · 2 GiB RAM)
  • Stop anytime — a stopped agent burns nothing
Buy credits — from $10
FAQ

Questions you should ask before paying anyone.

Why a microVM instead of a Docker container?
Docker shares one Linux kernel across every tenant. A bug in your neighbour's container can become your problem. Firecracker spins up a real KVM-backed virtual machine per tenant with hardware-enforced isolation — the same primitive AWS uses for Lambda.
What does BYOK actually mean here?
Your OpenRouter / OpenAI / Anthropic key lives only inside your VM. We never see it, never proxy, never mark up your spend. You pay your model provider direct.
What does the 30-day guarantee cover?
Cancel your first plan within 30 days for a full refund, no questions asked. The guarantee covers your first subscription and ends once you expand. After that, cancel anytime — your VM keeps running to the end of the period you've paid for, then simply doesn't renew.
Where do my VMs run?
AWS us-west-2 today, on EC2 nested-virtualization hosts — one Firecracker microVM per tenant. Additional regions land when committed regional MRR clears the cost of a second box.
Can I run multiple agents (Hermes + OpenClaw + Pi)?
Yes — each gets its own Firecracker microVM and its own subscription. So Hermes + OpenClaw + Pi at Starter is 3 × $25 = $75/mo. We deliberately don't bundle multiple harnesses inside one VM: it would break the per-tenant hardware-isolation boundary and starve resources at the Starter tier. Pricing is per-VM, so every agent you add is billed at its plan's rate — and annual prepay gives you 2 months free on each.
Can I really run a fleet of agents at once?
Yes — that's the swarm runtime. jurniti up --tier spot --harness hermes --count 50 --group demo boots fifty isolated microVM workers; jurniti run --group demo --model <model> --provider openrouter "<one brief>" dispatches the same brief to every worker (swarm slices carry no default model — you pick one per run, BYOK); jurniti cp --group demo work/out.txt ./out collects every result. Raise --count as your fleet grows — size is gated by your prepaid usage credits, not a plan cap, and workers bill by the second, so a stopped worker burns nothing.
What happens right after I pay?
Stripe confirms, you get a magic-link sign-in, and the auto-provisioner boots your microVM — live in about three minutes, no human in the loop. Paste your model key from the quickstart and your agent is answering a couple of minutes after that.
What are Memory and Observability?
Optional sidecars alongside your agents — same Firecracker isolation, not shared multi-tenant SaaS. Memory engines (Obsidian, gbrain, mem0) are live today: buy one, grant it over MCP, and agent knowledge compounds. Observability engines (Latitude, Langfuse) are the self-hosted trace catalog on the same attach model — sessions and tool calls stay in your microVM instead of a vendor SaaS. Neither stack is baked into every agent rootfs; you attach what you need.
Get started

Own your agents. Own the runtime.

From $21/mo billed yearly — or $25/mo. BYOK. 30-day money-back, refunds processed in one business day. Live in about three minutes.

See a working agent →

Annual · $250/yr · 2 months free · talk to us