Carina is an autonomous AI agent that runs locally, remembers everything, and gets smarter every session. Add Labyrinth Scout for enterprise-grade security monitoring, prompt injection protection, and real-time alerting.
A general autonomous agent with vertical depth no competitor ships by default. Free. Open. Yours.
Episodic memory with pgvector semantic search. Carina recalls relevant past conversations automatically.
After solving a complex problem, Carina writes a skill document. She gets smarter every session.
Web search, files, shell (sandboxed), email, HTTP requests, code execution. She acts, not advises.
DeepSeek, Claude, GPT-4o, or local Ollama. Type /provider anthropic — no restart.
CLI, Telegram, WhatsApp, Web UI. Same memory, same skills, same Carina.
Your data never leaves your machine unless you explicitly use a network tool. MIT licensed.
Carina Core is free. Labyrinth Scout adds the security control plane that enterprises require. Built from the ground up using deception technology from VERLOX's Master's distinction project.
Dynamically rotates Carina's API fingerprints. Attackers cannot reliably identify or target your agent. Port of the CMS technology that won a Master's distinction.
Three-layer detection: regex patterns, LLM-as-judge, and tool result scanning. Blocks direct and indirect injection attacks before they reach Carina's context.
Animated threat radar, live event stream, heat maps, and anomaly scoring. Every tool call, injection attempt, and anomaly appears within milliseconds.
Telegram, WhatsApp, Slack, Email, SMS. Get critical alerts instantly. Custom severity thresholds per channel.
Fake API keys, credential files, and endpoints that alert when accessed. Based on the Labyrinth deception model from your Master's project.
Learns Carina's normal behavior. Alerts on tool call frequency spikes, dangerous command patterns, and data exfiltration attempts.
Define which tools can run without confirmation. Rate limit per session. Block dangerous commands like `rm -rf /` or `curl | bash`.
Remotely suspend a compromised Carina instance from the dashboard. One click. Immediate effect via Redis pub/sub. Full audit trail.
Every action logged with cryptographic integrity. Export for GDPR, SOC2, ISO27001. Memory transparency — see and delete what Carina knows about you.
Block Carina from calling unknown domains. Detect and block sensitive data (API keys, passwords) from leaving your infrastructure.
Carina Core is MIT licensed and always free. Labyrinth Scout is the paid security control plane for production deployments.
npm install -g carina-agentFor any AI agent (OpenClaw, Hermes, AutoGPT, custom)
When monitoring Carina, the free open source agent. Saves £240/year.
Use Carina? Get 20% off →
Labyrinth Scout
For organisations running autonomous AI agents in production.
No per-seat pricing. No hidden fees. Cancel anytime.
Self-hosted Carina Core is always free. Scout is £99/mo with any agent, or £79/mo when paired with Carina. No telemetry. No hidden fees.
| Capability | OpenClaw | Hermes | Carina + Scout |
|---|---|---|---|
| Autonomous task execution | |||
| Persistent memory | |||
| Self-writing skills | No | ||
| Vertical skill packs (property-uk) | No | No | |
| Memory transparency (view/edit) | No | No | |
| Switch LLM mid-session | No | No | |
| npm install | No | No | |
| Prompt injection protection | No | Partial | 3-layer |
| Moving target deception | No | No | Scout |
| Real-time security dashboard | No | No | Scout |
| Kill switch | No | No | Scout |
| Open source (MIT) | |||
| Labyrinth Scout pricing | £99/mo | £99/mo | 20% off £79/mo |
One command. Runs on your machine or VPS. Your data stays yours.
Built by VERLOX Ltd · MIT License · carinaai.uk · labyrinthscout.com