Klyra

Klyra is a self-hosted AI security lab for authorized testing. It gives each operator a fresh LXD workspace, runs OpenCode inside that isolated container, and wraps the workflow in a browser UI with chat, files, sessions, admin controls, and clean findings.

Built for owned labs, internal appsec, CTFs, security education, and written-scope assessments. Do not use Klyra against systems you do not own or have explicit permission to test.

Why It Stands Out

• Fresh workspace on demand: user sessions are backed by LXD containers, not your host shell.
• OpenCode inside the container: no host binary leakage, no accidental local history reuse.
• Admin dashboard: create users, promote access, inspect containers, stop/delete workspaces.
• Evidence-first chat: streaming responses, command outputs, session records, and project files.
• Public-safe agent policy: the bundled agent policy is scoped to authorized testing and remediation.
• One-command demo deploy: Python dependencies, frontend build, LXD checks, backend start, and Cloudflare quick tunnel.

Architecture

Quick Start

Requirements:

• Ubuntu/Debian VPS or local Linux host
• Python 3.12+
• Node.js 18+
• LXD
• cloudflared optional for public demo tunnels

git clone https://github.com/CodexNexor/klyra.git
cd klyra
python3 deploy/run.py

The deploy script prints the local URL, optional public tunnel URL, and owner credentials. For production, set strong credentials before running:

export OWNER_USERNAME="owner"
export OWNER_PASSWORD="$(openssl rand -base64 32)"
export DB_MASTER_KEY="$(openssl rand -hex 32)"
python3 deploy/run.py

Common Commands

# Normal clean app start
python3 deploy/run.py

# Remove Klyra runtime state and app containers before deploy
python3 deploy/run.py --nuke

# Reinstall LXD from scratch, then deploy
python3 deploy/run.py --nuke-lxd

# LXD isolation smoke test
python3 isolation.py selftest

# Frontend build
cd frontend && npm install && npm run build

Admin Flow

1. Log in as the owner printed by deploy/run.py.
2. Open /admin.
3. Create a pro user or promote an existing user.
4. Open /playground.
5. Klyra provisions a fresh LXD container for that user and starts the chat workspace.

Security Model

Klyra is a lab boundary, not a magic trust boundary.

• The host runs the API, database, frontend, and LXD daemon.
• The operator workspace runs in an unprivileged LXD container.
• OpenCode is installed inside the container during provisioning.
• Session metadata is stored in SQLite; sensitive fields use application-level encryption.
• The bundled OPENCODE.md instructs agents to stay within authorized testing and avoid malware, credential theft, persistence, evasion, destructive actions, and unauthorized access.

Read SECURITY.md before exposing an instance.

Launch Checklist

• Set OWNER_PASSWORD and DB_MASTER_KEY.
• Enable host firewall rules.
• Restrict admin access to trusted operators.
• Use named Cloudflare tunnels or your own reverse proxy for production.
• Review OPENCODE.md for your organization’s policy.
• Configure log retention and backup policy.
• Test python3 isolation.py selftest.

Repository Topics

Suggested GitHub topics:

ai-security, appsec, pentest-lab, lxd, fastapi, react, opencode, security-automation, ctf, authorized-testing

Inspiration Notes

Popular AI-security repos in this space win attention with a short name, immediate screenshot/diagram, badges, a one-command demo, clear scope, and a strong safety/legal boundary. Klyra follows that launch shape while keeping the repo self-hosted and public-safe.

License

MIT. See LICENSE.