PowerAI Documentation
PowerAI is an invisible, local-first cognitive AI harness designed for modern terminals (macOS, Linux, and Windows).
Introduction
PowerAI transforms your standard terminal into an intelligent assistant without sending any keystrokes or environment data over the internet. Running on top of Ollama, it translates natural language requests into exact executable CLI commands, understands output buffers, and auto-corrects command typos on the fly.
All inferences occur locally on your machine via Ollama with Metal GPU (macOS) or CUDA acceleration (Linux/Windows). Zero telemetry is collected.
Installation
Install PowerAI in one step using the official automated installer for your platform:
macOS & Linux (Zsh / Bash)
Windows (PowerShell 5.1 / 7+)
Ollama Setup & Models
PowerAI connects to Ollama running locally on port 11434. If you haven't installed Ollama yet, download it from ollama.com.
Recommended lightweight local models:
Natural Language Queries
Prefix your request with ai followed by your query in natural language:
Output Memory Buffer
PowerAI captures the last terminal output lines automatically into a rolling session buffer. You can query printed data without copying or pasting:
Typo Auto-Healing Hook
When you mistype a command (e.g. mrdir or dockr ps), the native shell error handler intercepts the command and generates the corrected syntax with explanation.
CLI Reference
| Command | Description |
|---|---|
ai "<query>" |
Translates natural language to native shell commands. |
ai version / ai -v |
Displays installed version and checks GitHub for releases. |
ai update / ai upgrade |
Downloads, verifies, and installs the latest version instantly. |
Environment Variables
| Variable | Default | Description |
|---|---|---|
POWERAI_MODEL |
llama3.2 |
Ollama model to use for local inferences. |
POWERAI_HOST |
http://127.0.0.1:11434 |
Ollama local or remote API endpoint. |
POWERAI_MAX_TOKENS |
150 |
Maximum tokens allocated per response. |
Zero-Python Native Architecture
Unlike traditional AI CLI wrappers that require a heavy Python interpreter runtime, PowerAI uses a 100% native architecture:
- Unix (macOS / Linux): C-compiled
jqwith POSIX shell regex fallback. Latency < 2ms. - Windows: Native .NET / C# in PowerShell module without external runtime dependencies.
Troubleshooting & FAQ
Error: Connection refused (11434)
Ensure Ollama is running in the background. Start it by opening the Ollama application or running ollama serve in a separate terminal.