ppxai Documentation¶
Multi-LLM Interface for Developers - Use your favorite AI provider in terminal or VSCode with zero vendor lock-in.
Why ppxai?¶
| Problem | ppxai Solution |
|---|---|
| Locked to one AI vendor | Switch between Perplexity, Gemini, OpenAI, OpenRouter, Ollama anytime |
| Can't use local models | Full Ollama/vLLM support with same interface |
| AI modifies files without asking | Consent-based safety for all file operations |
| Lost context when switching models | Preserve conversation across provider changes |
| Expensive cloud-only pricing | Mix cloud and local models per task |
Quick Start¶
Installation¶
Configure API Keys¶
# Create ~/.ppxai/.env with your keys
PERPLEXITY_API_KEY=pplx-xxx
GEMINI_API_KEY=xxx
OPENAI_API_KEY=sk-xxx
Run¶
ppxai # Rich TUI (original)
ppxaide # Textual TUI (v1.15.0+)
ppxai-server # HTTP server for VSCode
ppxai-desktop # Desktop web app
Features¶
Multi-Provider Support¶
- Perplexity AI - Real-time web search with citations
- Google Gemini - 1M token context, search grounding
- OpenAI - GPT-4o, o1 reasoning models
- OpenRouter - 100+ models including Claude
- Local Models - Ollama, vLLM, LMStudio
Four Interfaces¶
| Feature | ppxai (Rich TUI) | ppxaide (Textual TUI) | VSCode | Web App |
|---|---|---|---|---|
| Streaming responses | ✅ | ✅ | ✅ | ✅ |
| Syntax highlighting | ✅ | ✅ | ✅ | ✅ |
| File editing tools | ✅ | ✅ | ✅ | ✅ |
| Agent mode | ✅ | ✅ | ✅ | ✅ |
| Checkpoint/undo | ✅ | ✅ | ✅ | ✅ |
| Markdown in chat | Limited | ✅ Full rendering | ✅ | ✅ |
| Themes | 6 themes | 17+ themes | N/A | N/A |
| Tab completion | ✅ | ✅ | N/A | N/A |
Agent Mode¶
Enable autonomous multi-step task execution:
The AI can chain tool calls, edit files, run commands - all with your consent.
Bootstrap Context¶
Load project-specific instructions from AGENTS.md:
---
provider_hints:
local:
- "Complete tasks fully without stopping."
model_hints:
"deepseek-r1*":
- "Show reasoning before actions."
---
# Project Instructions
Python 3.11+, pytest for testing.
Documentation¶
-
Installation
Install ppxai on Linux, macOS, or Windows
-
Agent Mode
Autonomous task execution with tools
-
Bootstrap Context
Project-specific AI instructions
-
Provider Setup
Configure AI providers