currently: building in public — launching Q3 2026

Your AI agent.
Managed. Secure.
Not your problem.

We built Nemi because self-hosting OpenClaw gave us trust issues. CVEs, malicious skills, $3,600 LLM bills, no RBAC, one framework. We fixed all of it.

Run OpenClaw, Moltis, and NemoClaw on hardened infrastructure. RBAC, SSO, Firecracker isolation, and LLM cost controls — all included. Ready in 60 seconds.

waitlist.shearly_access

✓firecracker_microvm_per_tenant.shisolation: kernel-level

✓llm_bundled_with_spending_caps.confopenrouter + ollama + byok — no surprise bills

✓egress_security.ciliumhost_metadata: deny_always — internet: open

✓rbac_sso_audit_log.yamlyour security team will approve this

✓skill_scanner_verified.jsonno ClawHub roulette

→multi_framework_support.rsOpenClaw + Moltis + NemoClaw

→cve_fleet_patches.cronhours, not days

138

OpenClaw CVEs in 63 days

341

malicious skills found on ClawHub

42k+

self-hosted instances exposed on Shodan

$3,600

max LLM bill from one loop bug

the problem

Self-hosting AI agents is a part-time job you didn't sign up for.

problem_01.md

CVEs land on your lap

OpenClaw ships 2.2 security vulnerabilities per day. CVSS 8.8 RCE. One malicious link, full machine compromise. You patch it manually or you're exposed indefinitely.

problem_02.md

Agent loops drain your wallet

A misconfigured skill or a bad prompt burns $3,600 in LLM credits overnight. No cap, no warning, no way back. You find out on your bank statement.

problem_03.md

Skills are a supply chain risk

341 malicious skills confirmed on ClawHub in one audit. They exfiltrate API keys, install backdoors, run cryptominers. The marketplace has no real vetting process.

problem_04.md

Your team can't use it

No RBAC. No SSO. No audit logs. Your security team blocks it. Your compliance team blocks it. It stays a solo hobby when it could run your entire operations.

problem_05.md

15+ hours before first message

Docker, Tailscale, UFW, WebSocket auth, SSL renewal, port mapping. Setup takes a weekend. Then you own every outage, every patch, every 3am incident forever.

problem_06.md

One framework, one point of failure

Every managed host is OpenClaw-only. OpenAI now controls the project. If it pivots, charges licensing, or gets shut down — your entire operation dies with it.

what nemi does differently

Everything fixed. Nothing compromised.

smart routing

Your agent picks the right model automatically.

Most platforms forward every request to the same LLM. Nemi's built-in smart router classifies each task and selects the optimal model — cheap for simple queries, premium only when reasoning demands it. Most tenants see 60–80% lower API costs vs direct API usage.

1classify

Task classification

Each request is classified by type: simple Q&A, complex reasoning, code generation, long context, or image understanding.

2route

Optimal model selection

Simple tasks route to cost-efficient models. Complex reasoning gets premium models. Your Ollama endpoint gets priority when available.

3cap

Budget-aware decisions

As your spending cap approaches, the router automatically downgrades to cheaper models. No surprises. No overages. Real-time balance visible in dashboard.

smart_router.conf

# smart router decision
incoming_request:
  simple_qa       → kimi-k2.5        # $0.002/req
  reasoning       → claude-haiku      # $0.01/req
  code_generation → qwen-coder        # $0.004/req
  long_context    → gemini-pro        # $0.008/req

spending_cap: $100/mo
remaining: $73.20
strategy: smart           # smart | cheapest | specific
ollama_first: true        # $0 when available

security model

Built for the paranoid. Approved by compliance.

🔥

Firecracker isolation

Every agent runs in its own MicroVM with a dedicated kernel. A compromised skill can't escape. Cross-tenant attacks are structurally impossible — not just unlikely.

🔐

Secrets never touch the agent

API keys live in a Go sidecar proxy. Your agent only sees localhost. If a malicious skill takes over, there are no credentials to steal. Nothing in memory, nothing in env vars.

🛡️

CVE patches in hours, not days

138 OpenClaw CVEs in 63 days. We monitor, validate in staging, and roll patches across your fleet automatically. You don't lift a finger.

comparison

Why switch from where you are now.

feature	nemi	blink claw	clawctl	kiwiclaw	self-hosted
Firecracker MicroVM	yes — all paid tiers	no	no	no	manual
LLM bundled (200+ models)	yes — Pro+	yes	no — BYOK only	$39 tier only	no
Spending caps + real-time balance	yes	no	no	no	no
API key never exposed to skills	yes — sidecar proxy	no	no	no	no
RBAC + SSO (SAML/OIDC)	yes — Teams+	no	yes	no	no
Audit log (exportable)	yes — up to 1yr	no	yes	no	no
Curated skill scanner	yes + sandbox	basic	yes	partial	no
Multi-framework support	yes — 3 frameworks	no	no	no	manual
CVE fleet patching	hours, automatic	hours	manual	manual	your problem
Ollama / custom LLM endpoint	yes — Pro+	no	no	no	manual
Host metadata blocked	yes — enforced always	unknown	unknown	unknown	manual
Subdomain per tenant	yourdomain.nemi.run	no	no	no	manual
Starting price	$29/mo	$45/mo	$49/mo	$15/mo	$5/mo + your time

roadmap

Where we're going.

phase_01

Q3 2026

building now

Core platform

OpenClaw on Firecracker MicroVM. LLM bundled with spending caps and sidecar proxy. Skill scanner Level 1. Subdomain per tenant. Automatic CVE fleet patches.

phase_02

Q4 2026

coming soon

Teams and compliance

RBAC with 3 roles. SSO via Google, GitHub, SAML and OIDC. Exportable audit logs up to 1 year. Moltis framework support. Skill scanner with behavioral sandbox. SOC 2 Type I process begins.

phase_03

Q1 2027

planned

Enterprise and Nemi Shield

NemoClaw support. Visual multi-agent workflow builder. SOC 2 Type I complete. Nemi Shield — skill scanning as a public API for other platforms. Second region for APAC.

how it works

From signup to running agent in 60 seconds.

Sign up and pick a plan

Create your account, choose Solo, Pro, or Teams. 14-day trial with card. Waitlist members get 3 months free.

Configure your agent

Pick your framework, connect your LLM (bundled, BYOK, or Ollama), import your skills, and set your spending cap.

You're live

Your agent runs on Firecracker-isolated infrastructure with automatic CVE patches, daily backups, and a private subdomain. You ship features — we handle ops.

✓SOC 2 Type I in progress✓AWS infrastructure✓Cloudflare protected✓GDPR ready

faq

Questions you're already asking.

Can I migrate from my self-hosted OpenClaw?

Yes. Our migration assistant imports your backup — memory, skills, channels, and config. Zero downtime, no re-pairing.

What happens if OpenClaw gets shut down or pivots?

We support three frameworks: OpenClaw now, Moltis in Phase 2, NemoClaw in Phase 3. If one pivots, you migrate in minutes.

Where is my data stored?

AWS us-east-1 with encrypted EFS storage, per-tenant isolation, and daily backups with 30-day retention. EU region coming in Phase 2.

Do you have access to my conversations?

No. Your agent runs in an isolated MicroVM. We manage infrastructure, not your data. Secrets live in AWS Secrets Manager with 7-day rotation — we never see them.

What if an update breaks my agent?

Every update goes through staging first. If it breaks your config, we roll back automatically within 5 minutes — before you notice.

How does the smart router save money?

It classifies each task and routes simple queries to cheap models, using premium models only for complex reasoning. Most tenants see 60–80% lower API costs.

Stop self-hosting.
Start shipping.

Join the waitlist. Early access members get 3 months free and direct input into what we build first. No credit card until we launch.

Your AI agent.Managed. Secure.Not your problem.