jurniti

Product Introduction

1. Definition: jurniti (https://www.jurniti.com) is a secure managed AI agent platform and private always-on box for AI agents. Its technical category is Firecracker microVM hosting for agent runtimes: each agent runs in its own Firecracker microVM with a KVM hardware boundary, the same virtualization primitive AWS uses for Lambda, not a shared Docker container. Operated by OranAI LLC dba jurniti, the platform provisions, bills, and manages agent microVMs through a developer CLI, MCP server, REST API, and browser terminal. It is not a model API, prompt proxy, or managed agent fork; it ships upstream harness installers at first boot.

2. Core Value Proposition: jurniti exists to run always-on AI agents in real isolated computers with hardware-enforced tenant separation, BYOK model key isolation, persistent disk, in-browser terminal, public TLS subdomain, and forkable agent templates. Primary keywords include managed AI agent hosting, Firecracker microVM hosting, KVM-isolated agent VM, BYOK AI agent platform, always-on Claude Code server, Codex CLI hosting, Hermes Agent hosting, OpenClaw hosting, Pi coding agent hosting, and agent template forking. Pricing starts at $25/mo per VM with a 30-day money-back guarantee on the first purchase; there is no free trial.

Main Features

1. Firecracker microVM per agent with KVM hardware isolation: Each agent receives its own Firecracker microVM, not a shared-kernel Docker container. Firecracker is the same microVM primitive AWS uses for Lambda, and jurniti enforces a KVM hardware boundary per tenant. VMs run on AWS EC2 nested-virtualization hosts, initially in us-west-2, with one density-sliced host per region; additional regions launch when per-region MRR covers the cost of a second box. Firecracker boots in about 124ms, while full provisioning from Stripe checkout to a fresh shell terminal takes about 3 minutes. CPU and RAM are dedicated through pinned cgroup reservations, not noisy-neighbor slices. This architecture is designed for hardware-enforced isolation, multi-tenant security, and predictable agent performance.

2. BYOK model key isolation and no token proxy: jurniti uses bring-your-own-key (BYOK) for model access. Your model API keys live only inside your VM guest, and jurniti never sees, proxies, or marks up your spend. This supports OpenRouter, Nous Portal, OpenAI, Anthropic, and other provider keys depending on the harness. The platform is not a model API, prompt proxy, or usage-metered LLM gateway. BYOK key isolation is central to the product’s security posture, cost control, and compliance story for teams that cannot expose prompts, tokens, or provider credentials to a third-party proxy.

3. Developer CLI, MCP server, and REST API: Install the signed developer CLI on macOS or Linux with curl -fsSL https://jurniti.com/install.sh | sh, then run jurniti login, jurniti billing add-card, and jurniti up. The verb map includes jurniti login, whoami, harnesses, up, run, cp, auth, stop, start, restart, reprovision, capture, publish, fork, vms ls, vms get, templates ls, usage, billing, keys, env, skills, docs, and mcp. Mount jurniti as an MCP server for your own agent with claude mcp add jurniti -- jurniti mcp. The same binary exposes stdio MCP, and the CLI verb equals the REST noun equals the MCP tool. The REST API uses a customer key prefixed jrn_live_, with OpenAPI at https://www.jurniti.com/openapi.json and host https://api.jurniti.com. Errors return {error, message}. This gives agents and CI systems one unified control plane for provisioning, forking, and driving isolated agent VMs.

4. Broad harness support for upstream agent runtimes: jurniti supports Claude Code, Codex, Hermes, OpenClaw, Pi, Claude Desktop, Cursor Agent, Devin, DeepAgents Code, OpenCode, Goose, Grok Build, fx, Gemini CLI, Cline, Kilo Code, GitHub Copilot CLI, OpenHands, Prime Agent, Aider, Amp, Pool, Paperclip, Mastra, OpenScience, Hermes Desktop, Unsloth Desktop, QM, Headlong, and Exo. Each harness has its own landing page and .md twin, with the canonical always-current list at https://www.jurniti.com/llms-full.txt. jurniti ships each harness’s upstream installer at first boot, with no proprietary patches. Each harness gets its own subscription, its own dedicated microVM, and its own dedicated CPU plus RAM. Running Hermes, OpenClaw, and Pi at Starter is 3 × $25 = $75/mo.

5. Persistent disk, in-browser terminal, and public TLS subdomain: Every agent VM includes a persistent disk, an in-browser terminal, and a subdomain. Pro plans add a custom subdomain, and Max plans add daily snapshots. Persistent storage keeps home directories, workspaces, agent memory, and configuration alive across restarts. Public TLS subdomains make agent dashboards and bridges reachable, such as a Hermes Dashboard at yourname.jurniti.com or OpenClaw bridges for Discord, WhatsApp, Signal, and Telegram. Pi runs with a persistent workspace and write access to extensions the agent installs itself. These features make jurniti suitable for always-on coding agents, messaging bridges, and web-accessible agent control surfaces.

6. Templates and forkable agent behavior: A jurniti template is a snapshot of a configured agent VM’s persist disk, captured by its owner and published so anyone can fork it into their own new VM. The model is classic GitHub: public templates are unlimited and free forever, while private templates use a count-based Templates plan. Private tiers are t5, t10, t20, and t50 for 5, 10, 20, and 50 private slots at $5, $9, $17, and $29 per month, with additive 50-slot blocks past t50; there is no free private tier. Secrets and PII are scrubbed on public publish, and a fork boots a fresh VM seeded with the template’s files and config while the user brings their own model keys. Forking starts a normal Stripe Checkout for a new VM at the same flat plan price, with no fork fee, and auto-provisions a fresh microVM after payment confirms. This turns agent configuration into a shareable, forkable artifact.

7. Swarm runtime and sidecars for multi-agent fleets: jurniti supports parallel swarms with commands such as jurniti up --tier spot --count N followed by jurniti run and jurniti cp. Each agent in a swarm gets its own KVM boundary, which is important for isolation when running many autonomous agents. The platform also includes memory and observability sidecars, plus an agent org planner tool for mapping planner, worker, messenger, and researcher roles to isolated microVMs, one microVM per role. This supports agentmaxxing, the practice of running many agents in parallel, and gives AI-native companies a runtime model for multi-agent fleets with per-agent isolation, observability, memory, and communication channels.

Problems Solved

1. Pain Point: Shared Docker containers and shared-kernel sandboxes do not provide hardware-enforced tenant isolation, which creates security and noisy-neighbor risks for autonomous agents. Laptop-bound agents stop when the lid closes, so always-on Claude Code, Codex, Hermes, OpenClaw, or Pi workflows are unreliable. Sending model API keys to a proxy platform exposes credentials, prompts, tokens, and spend to a third party, while token-markup gateways increase cost and reduce control. DIY VPS hosting adds ops overhead for patching, networking, TLS, persistence, provisioning, and agent lifecycle management. Teams also need a repeatable way to fork, share, and version agent behavior without hand-building every environment. jurniti addresses these managed AI agent hosting, BYOK security, always-on agent server, and Firecracker isolation problems.

2. Target Audience: AI engineers and agent developers building with Claude Code, Codex, Hermes, OpenClaw, Pi, Gemini CLI, Cline, Aider, OpenHands, and other harnesses. DevOps and platform engineers who need isolated agent VMs with CLI, MCP, and REST automation. Technical founders and indie hackers shipping AI-native products without maintaining VPS fleets. Forward Deployed Engineers (FDEs) and automation consultants deploying always-on agent fleets for clients. Research teams running parallel agents for data collection, analysis, and experimentation. Growth and support operations teams using OpenClaw bridges for Discord, WhatsApp, Signal, and Telegram. Security-conscious organizations that require BYOK key isolation and hardware-enforced KVM boundaries. Teams comparing E2B, Railway, Replit, Modal, GitHub Codespaces, and other sandbox or VPS hosts for agent workloads.

3. Use Cases: Always-on Claude Code on a dedicated microVM so coding agents keep running after the laptop closes. Codex CLI hosting for persistent terminal-based coding workflows. Self-hosted Hermes Agent with BYOK for OpenRouter, Nous Portal, OpenAI, or Anthropic keys, plus a persistent home directory and dashboard at a jurniti subdomain. Managed OpenClaw hosting that bridges Discord, WhatsApp, Signal, or Telegram to an LLM using the existing plugin ecosystem. Pi coding agent hosting with a 4-tool core, self-extending behavior, JS-native runtime, and write access to installed extensions. Forking a public template into a new paid microVM to reproduce a configured agent environment. Running parallel swarms with jurniti up --tier spot --count N for research, scraping, testing, or fleet dispatch. Isolating dangerous permissions such as --dangerously-skip-permissions inside a VM instead of a laptop. Deploying multi-agent orgs with planner, worker, messenger, and researcher roles on separate microVMs. Provisioning from CI or an external agent through the REST API, MCP server, or developer CLI.

Unique Advantages

1. Differentiation: jurniti competes with shared-container sandboxes, VPS hosts, model proxies, and DIY agent infrastructure by offering one Firecracker microVM per agent with KVM hardware isolation. Unlike E2B, Railway, Replit, Modal, or GitHub Codespaces, jurniti’s wedge is real hardware isolation for each tenant and BYOK model keys that stay inside the guest VM. Unlike model gateways, jurniti does not see, proxy, or mark up token spend. Unlike a plain VPS, jurniti provides a managed control plane with CLI, MCP, REST, templates, forking, persistent disk, in-browser terminal, subdomain, snapshots, and flat monthly pricing. Public templates are unlimited and free forever, while private templates use count-based plans. Every comparison page includes a mandatory “where they’re strong” honesty block and a “pick them when…” line, which supports fair evaluation rather than one-sided marketing.

2. Key Innovation: The key innovation is the combination of Firecracker microVM isolation with a unified agent control plane. Firecracker is the same primitive AWS uses for Lambda, and jurniti applies it per agent with a KVM hardware boundary, ~124ms boot, and ~3 minute full provisioning. BYOK keeps model keys inside the VM, so the platform never sees prompts, tokens, or spend. The CLI verb, REST noun, and MCP tool are one operation, letting humans, CI, and agents provision and drive VMs through the same surface. Templates turn a configured agent VM into a forkable artifact with public unlimited sharing and paid private slots. Swarm runtime, sidecars, persistent disks, subdomains, and snapshots complete the model for always-on, isolated, multi-agent infrastructure.

Frequently Asked Questions (FAQ)

1. What is jurniti? jurniti is a secure managed AI agent platform that runs each agent in its own Firecracker microVM with a KVM hardware boundary. It is not a shared Docker container, model API, prompt proxy, or managed agent fork. It provides always-on agent hosting with BYOK key isolation, persistent disk, in-browser terminal, public TLS subdomain, CLI, MCP, REST API, templates, forking, and swarm support. Pricing starts at $25/mo per VM with a 30-day money-back guarantee and no free trial.

2. How is jurniti different from a Docker container or VPS? A shared Docker container uses a shared kernel and does not provide a KVM hardware boundary per tenant. A raw VPS gives isolation but leaves provisioning, networking, TLS, persistence, snapshots, agent lifecycle, and template forking to the user. jurniti combines Firecracker microVM isolation, the same primitive AWS uses for Lambda, with a managed control plane for CLI, MCP, REST, templates, and billing. Each agent gets its own microVM, dedicated CPU and RAM via pinned cgroup reservations, persistent disk, and subdomain.

3. Does jurniti support BYOK and keep my API keys private? Yes. jurniti is built around bring-your-own-key model access. Your model API keys live only inside your VM guest, and jurniti never sees, proxies, or marks up your spend. Depending on the harness, you can use keys for OpenRouter, Nous Portal, OpenAI, Anthropic, and other providers. This makes jurniti suitable for teams that require BYOK AI agent hosting, credential isolation, and no token-markup gateway.

4. How much does jurniti cost, and is there a free trial? jurniti pricing is flat per VM, billed monthly or annual with two months free on annual. Starter is $25/mo or $250/yr with 1 vCPU, 2 GiB RAM, 10 GiB SSD, and 100 Mbit/s. Pro is $49/mo or $490/yr with 2 vCPU, 6 GiB RAM, 50 GiB SSD, 200 Mbit/s, and a custom subdomain. Max is $99/mo or $990/yr with 4 vCPU, 12 GiB RAM, 100 GiB SSD, 500 Mbit/s, and daily snapshot. There is no free trial and no free tier; the de-risk is a 30-day money-back guarantee on the first purchase. Running multiple harnesses means multiple subscriptions, so Hermes plus OpenClaw plus Pi at Starter is 3 × $25 = $75/mo.

5. Can I run Claude Code, Codex, Hermes, OpenClaw, or Pi on jurniti? Yes. jurniti supports Claude Code, Codex, Hermes, OpenClaw, Pi, Claude Desktop, Cursor Agent, Devin, DeepAgents Code, OpenCode, Goose, Grok Build, fx, Gemini CLI, Cline, Kilo Code, GitHub Copilot CLI, OpenHands, Prime Agent, Aider, Amp, Pool, Paperclip, Mastra, OpenScience, Hermes Desktop, Unsloth Desktop, QM, Headlong, and Exo. Each harness runs in its own microVM, and jurniti ships the upstream installer at first boot with no proprietary patches. You can provision from the CLI, mount jurniti as an MCP server, or call the REST API.