AI Terminal Copilot
for UNIX/Linux

Explain commands. Suggest fixes. Review before you execute.
Self-hosted. Inspectable. No cloud required.

Try the Demo

Built for people who live in the terminal

$_

Sysadmins

Debug services, inspect logs, fix permissions — with AI that understands UNIX.

⚙

DevOps / SRE

Docker, systemd, nginx, PostgreSQL — get correct commands, not Stack Overflow guesses.

🔒

Security Teams

Every command is safety-classified before execution. Dangerous actions require confirmation.

⌨

Platform Teams

Self-hosted, API-first. Integrate into internal tools, runbooks, or incident workflows.

What it does

/ask

Ask any UNIX/Linux question. Get a clear, concise answer with real commands.

/suggest

Describe what you want to do. Get a safe command with an explanation and risk level.

/review

Paste a command before running it. Get a safety classification and step-by-step breakdown.

/analyze

Paste command output, logs, or errors. Get a diagnosis and suggested fixes.

Safety Layer

Deterministic, rule-based command classifier. Not a prompt trick — actual pattern matching.

Knowledge Grounding

Mount your docs, runbooks, or man pages via aifs. Answers are grounded in real documentation, not just LLM training data.

Session History

Every interaction is logged locally. Audit what was asked, suggested, and executed.

Example Use Cases

unixpilot ask "Why is port 443 showing TIME_WAIT?"

Explains TCP states and suggests diagnostic commands.

unixpilot suggest "find large files over 1GB in /var"

Returns: find /var -type f -size +1G -exec ls -lh {} \; with safety: safe.

unixpilot review "rm -rf /tmp/*"

Flags as DANGEROUS — recursive forced deletion. Requires confirmation.

unixpilot analyze < error.log

Reads error output, identifies the failure, suggests a fix.

unixpilot suggest "restart nginx after config change"

Returns: nginx -t && systemctl reload nginx — tests config first.

unixpilot ask "How do I add a swap file on Debian?"

Step-by-step: fallocate, mkswap, swapon, fstab entry.

Architecture & Trust Model

┌─────────────┐     ┌──────────────┐     ┌──────────────┐     ┌─────────────┐
│  CLI / API  │────▶│  Safety      │────▶│  Knowledge   │────▶│  AI Engine  │
│  Client     │     │  Classifier  │     │  (aifs)      │     │  (LLM)      │
└─────────────┘     └──────┬───────┘     └──────────────┘     └──────┬──────┘
                           │                                         │
                    ┌──────▼───────┐                          ┌──────▼──────┐
                    │  Execution   │                          │  Response   │
                    │  Gate        │                          │  Formatter  │
                    └──────────────┘                          └─────────────┘

Trust model

  • Safety classifier runs first — before the AI, before execution. It's rule-based and deterministic.
  • Execution is off by default. Commands are never auto-run. You opt in explicitly.
  • Self-hosted. Your queries stay on your infrastructure. Use Ollama, vLLM, or any OpenAI-compatible backend.
  • Auditable. Every interaction is logged to local JSONL files.
  • Knowledge-grounded. Optionally mount docs via aifs — answers cite real documentation, not just training data.
  • No hidden prompts. The system prompt is in the source code. What you see is what runs.

Live Demo

Try it live — powered by a self-hosted LLM on our infrastructure.