Security by architecture

Your AI assistant.
Your own server.

Every customer gets a dedicated VPS — not a container, not a shared pool. No cross-tenant risk. Smart model routing across 30+ LLMs built in.

30+
Models
0
Shared servers
443
Port only
01

Dedicated infrastructure

Your machine. Nobody else's.

Most AI platforms run all customers on shared servers. One breach affects everyone. Your Telsi instance is a separate machine — provisioned on Hetzner Cloud, owned by your account, torn down when you leave.

Not a container. Not a shared pool. A real VPS.
Shared AI platform
Thousands of tenants, one pool
Your Telsi instance
Dedicated VPS, zero neighbors
02

Hardened by default

Security controls you can audit.

Every instance ships with the same baseline: no SSH, no root, no open ports except 443. Credentials never touch disk unencrypted. Cloud metadata service blocked at the network layer.

Every CVE patched before deployment
iptables firewall
Port 443 only — no SSH
Non-root process
openclaw runs as unprivileged user
systemd sandbox
PrivateTmp, NoNewPrivileges, CapabilityBoundingSet
AES-256-GCM
Bot tokens encrypted at rest
Cloudflare Full Strict
End-to-end TLS, no half-cert
Metadata blocked
169.254.169.254 unreachable
03

Smart model routing

Right model, right cost.

Every message is classified locally in under 1ms — no API call. Simple questions route to fast, cheap models. Complex analysis routes to powerful ones. You get the quality you need at the price you deserve.

70-85% cost savings vs single-model approach
SIMPLE"What is 2+2?"
DeepSeeksaved 99%
MEDIUM"Summarize this article"
GPT-4o-minisaved 95%
COMPLEX"Build a React component"
Claude Sonnetsaved 80%
REASONING"Prove this theorem"
Claude Opus
04

The routing engine. 30+ models, one subscription.

Every message is classified locally in under 1ms — no API call. Simple questions route to fast, cheap models. Complex analysis routes to powerful ones. ClawRouter handles all of it automatically.

With routing
$60–180
underlying model cost / mo
Without routing
$450–900
single premium model
70-85% savings
SIMPLE~60%
Gemini 2.5 Flash
Greetings, yes/no, lookups
Cost / M tokens
$0.15 / $0.60
MEDIUM~25%
DeepSeek V3
Writing, summaries, code
Cost / M tokens
$0.28 / $0.42
COMPLEX~10%
Claude Sonnet 4
Analysis, strategy, research
Cost / M tokens
$3.00 / $15.00
REASONING~5%
Claude Opus 4.5
Legal, math, contracts
Cost / M tokens
$15.00 / $75.00
Launch pricing

Simple pricing. Dedicated VPS on every tier.

Every plan includes your own server. No shared infrastructure. No hidden costs.

Starter

Your own VPS, 30+ models, ready in 30 seconds

$49
$29/mo
Get started
  • 5,000 messages/month
  • 30+ AI models
  • Dedicated VPS
  • 443-only firewall
  • AES-256-GCM encryption
Most popular

Pro

For professionals who rely on their assistant daily

$99
$79/mo
Get started
  • 25,000 messages/month
  • 30+ AI models
  • Dedicated VPS
  • Priority support
  • 443-only firewall
  • AES-256-GCM encryption

Business

Full platform for teams and power users

$199
$149/mo
Get started
  • 100,000 messages/month
  • 30+ AI models
  • Dedicated VPS
  • Priority support
  • Custom skills
  • 443-only firewall
  • AES-256-GCM encryption

Your AI assistant. Provisioned in 30 seconds.

Sign up, connect your Telegram bot, and start chatting. Your dedicated instance is ready before you finish reading this.

Dedicated VPSNo shared infrastructureCancel anytime