Uatu

AI-powered system troubleshooting agent using Claude.

What it does:

• Chat with your system to diagnose issues
• Pipe logs directly for instant analysis
• Approve commands before execution with granular controls
• Connects symptoms across CPU, memory, network, and processes

Platforms:

• macOS
• Linux

Installation

Using pipx (recommended)

# Install with pipx for isolated environment
pipx install uatu

# Configure API key
echo "ANTHROPIC_API_KEY=your_key" > .env

Using pip

# Install globally or in a virtual environment
pip install uatu

# Configure API key
echo "ANTHROPIC_API_KEY=your_key" > .env

From source with uv

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Clone and install
git clone https://github.com/fractalops/uatu.git
cd uatu
uv sync

# Configure API key
echo "ANTHROPIC_API_KEY=your_key" > .env

Quick Start

Interactive Chat

Start a troubleshooting session:

# Read-only mode (default) - uses MCP tools, no bash
uatu

# Allow bash commands with approval prompts
UATU_READ_ONLY=false uatu

Example questions:

• "What's causing high CPU usage?"
• "Why is my server running slowly?"
• "Investigate recent memory issues"
• "What's listening on port 8080?"

Commands require approval unless allowlisted.

Stdin Mode

Pipe data for analysis:

# Read-only (default) - MCP tools only
cat /var/log/app.log | uatu "find errors"

# Enable bash commands (requires UATU_READ_ONLY=false)
UATU_READ_ONLY=false journalctl -u myservice | uatu "why did this crash?"

# Safe bash commands (ps, df, top, etc.) auto-approve from allowlist
UATU_READ_ONLY=false ps aux | uatu "diagnose memory issues"

Allowlisted commands auto-approve in stdin mode. TTY required for interactive approval prompts.

Configuration

Create .env:

# Required
ANTHROPIC_API_KEY=your_key

# Security (defaults shown)
UATU_READ_ONLY=true                     # true: MCP tools only, false: allow bash
UATU_REQUIRE_APPROVAL=true              # Require approval before bash execution
UATU_ALLOW_NETWORK=false                # Allow network access (WebFetch, WebSearch)

# Agent features
UATU_MODEL=claude-sonnet-4-5-20250929  # Claude model
UATU_ENABLE_SUBAGENTS=true              # Specialized diagnostic agents
UATU_ENABLE_SKILLS=true                 # Filesystem-based agent skills
UATU_MAX_BACKGROUND_JOBS=2              # Concurrent background bash jobs

# UI options
UATU_CONSOLE_WIDTH=80                   # Terminal width (80, 0=full, None=auto)
UATU_SHOW_TOOL_PREVIEWS=true            # Show tool result previews
UATU_SHOW_STATS=true                    # Show session token/cost stats

# Telemetry (opt-out)
UATU_ENABLE_TELEMETRY=true              # Local session telemetry
UATU_TELEMETRY_PATH=~/.uatu/telemetry.jsonl

Security

Command Approval

Bash commands require approval unless allowlisted:

⚠ Bash command approval required
Risk: Credential Access

⚠ Warning: This command may access SSH keys, certificates, or other credentials

ls -la ~/.ssh/

  ○ Allow once
  ○ Always allow (exact)
  → Deny

Audit Log

Security decisions are logged to ~/.config/uatu/audit.jsonl:

# View audit log (last 100 events)
uatu audit

# View recent events
uatu audit --last 20

# View specific event types
uatu audit --type bash_command_approval

# View summary statistics
uatu audit --summary

Allowlist

Manage approved commands in ~/.config/uatu/allowlist.json:

# View directly
cat ~/.config/uatu/allowlist.json

# In interactive chat (with tab completion)
/allowlist                    # Show approved commands
/allowlist add <command>      # Add to allowlist
/allowlist remove <pattern>   # Remove from allowlist
/allowlist clear              # Clear all

Development

# Run tests
uv run pytest

# Lint
uv run ruff check .

# Format
uv run ruff format .

License

This project is licensed under the MIT License - see the LICENSE file for details.

Built With

• Claude Agent SDK for Python
• Rich - Terminal formatting
• Typer - CLI framework
• prompt_toolkit - Interactive input