Benchmarks

Security and performance measurements for ZYBER infrastructure.

Security Metrics

Data Exposure Reduction

Comparison of data points exposed to AI providers:

Data Point
Direct AI Use
Through ZYBER
Reduction

Real IP address

✅ Exposed

❌ Hidden

100%

Device fingerprint

✅ Exposed

❌ Hidden

100%

Browser cookies

✅ Accessible

❌ Isolated

100%

Local file system

✅ Accessible

❌ Isolated

100%

Email/identity

✅ Required

❌ Not required

100%

Geographic location

✅ Via IP

❌ Datacenter only

100%

OS/hardware info

✅ Exposed

❌ VM only

100%

Overall data exposure reduction: 75-90% depending on use case.

Attack Surface Comparison

Attack Vector
Local AI Tools
ZYBER

Malicious code execution

Direct device access

Workspace VM only

Credential theft

Access to credential stores

Workspace credentials only

Lateral movement

Full network access

No LAN access

Persistence

Can modify system

Ephemeral option

Privilege escalation

Possible

Contained to VM

Isolation Boundary Tests

Test
Result

VM escape attempts

0 successful (tested quarterly)

Cross-workspace access

0 successful

Host filesystem access

0 successful

Network namespace breakout

0 successful

Kernel exploit (container)

Mitigated by VM isolation

Performance Benchmarks

Workspace Startup Time

Operation
Time (p50)
Time (p99)

Ephemeral workspace creation

4.2s

8.1s

Persistent workspace resume

2.1s

4.5s

First terminal ready

5.8s

10.2s

Network Latency

Route
Latency Added

Workspace → AI API (direct)

+12ms avg

Workspace → GitHub

+8ms avg

Workspace → npm registry

+15ms avg

User → Workspace (stream)

+25-50ms (varies by region)

Throughput

Metric
Value

Workspace egress bandwidth

100 Mbps sustained

File upload to workspace

50 Mbps avg

File download from workspace

80 Mbps avg

Git clone speed

~90% of direct speed

Resource Performance

Resource
Basic Tier
Standard Tier
Performance Tier

CPU (Geekbench 6 single)

~1,200

~1,400

~1,600

CPU (Geekbench 6 multi)

~2,400

~5,200

~11,000

Disk I/O (sequential)

200 MB/s

400 MB/s

800 MB/s

Disk I/O (random 4K)

15K IOPS

30K IOPS

60K IOPS

Memory bandwidth

20 GB/s

25 GB/s

35 GB/s

Uptime & Reliability

Service Availability

Period
Uptime

Last 30 days

99.92%

Last 90 days

99.87%

Last 12 months

99.85%

Incident Response

Metric
Target
Actual

Time to detect

< 5 min

2.3 min avg

Time to respond

< 15 min

8.7 min avg

Time to resolve (P1)

< 1 hour

34 min avg

Time to resolve (P2)

< 4 hours

2.1 hours avg

Compliance & Audits

Third-Party Assessments

Assessment
Status
Last Completed

Penetration testing

✅ Passed

Q4 2024

Infrastructure audit

✅ Passed

Q4 2024

Code security review

✅ Passed

Q3 2024

Dependency audit

✅ Passed

Monthly

Security Controls

Control
Implementation

Encryption at rest

AES-256

Encryption in transit

TLS 1.3

Key management

Hardware Security Modules

Access logging

All API calls logged

Audit trail retention

12 months

Comparison: Local vs ZYBER

Security Posture

Category
Local Development
ZYBER
Improvement

Code execution isolation

None

Full VM

Credential exposure

High

Low

80% reduction

Network exposure

Full LAN

None

100% reduction

Persistence risk

High

Ephemeral option

100% reduction

Identity exposure

Full

Wallet only

90% reduction

Development Workflow Impact

Task
Local Time
ZYBER Time
Overhead

Clone medium repo

45s

48s

+7%

npm install (500 deps)

60s

65s

+8%

Build TypeScript project

12s

13s

+8%

Run test suite

30s

31s

+3%

AI code generation

2s

2.1s

+5%

Average overhead: 5-10%, acceptable tradeoff for security isolation.

Methodology

How We Measure

Metric
Method

Startup time

Automated synthetic tests, 1000 samples/day

Latency

Continuous monitoring from multiple regions

Throughput

iperf3 tests between components

Security tests

Quarterly penetration testing by third party

Uptime

External monitoring (Datadog, Pingdom)

Test Environment

  • Benchmarks run on production infrastructure

  • Results represent real user experience

  • Updated quarterly with latest data

  • Anomalies investigated and explained

Transparency

We publish:

  • Monthly uptime reports

  • Quarterly security assessment summaries

  • Incident post-mortems (when applicable)

Available at: status.zyber.app

Last updated