A versatile Python CLI for interacting with various Large Language Models (LLMs) from multiple inference providers. Pure stdlib, Python 3.9+ — zero pip installs required.

The original bash script (ai.sh) is also included for minimal environments.

• 🌐 Multi-Provider Support: Chat with models from:

  • Google Gemini
  • OpenRouter (access to multiple model providers)
  • Groq
  • Together AI
  • Cerebras AI
  • Novita AI
  • Ollama (local or cloud)

• 🎯 Dynamic Model Selection: Fetches and lists available models from the chosen provider with optional filtering.

• 💾 Conversation History: Remembers the last N messages (configurable). Use /history to recall conversation logs.

• 📝 System Prompt: Define a system-level instruction for the AI (configurable).

• ⚡ Streaming Responses: AI responses are streamed token by token for a real-time feel.

• 🎨 Color-Coded Output: Differentiates between user input, AI responses, thinking, tools, errors, and info messages.

• 📖 Markdown Rendering: Bold, italic, inline code, fenced code blocks — rendered beautifully in terminal.

• 🔢 LaTeX Rendering: Greek letters, superscripts, subscripts, fractions → Unicode (e.g., α², π/2).

• 🖼️ Vision Support: Attach images for vision-capable models.

  • /upload <path> — Attach an image
  • /image — Show attached image info
  • /clearimage — Remove attached image

• 🔄 Live Model Switching: Switch models mid-conversation with /model (preserves history).

• 🛠️ Tool/Function Calling: When enabled, the model can invoke local tools:

  • get_time — Current local date & time
  • calculator — Safe arithmetic evaluation
  • fetch_url — Fetch & clean a web page (≤ 8,000 chars)
  • wikipedia — Search Wikipedia & return article text
  • Gemini also gets Google Search grounding when tools are on.

• 🧠 Thinking/Reasoning Output: Toggle display of reasoning tokens from supported models with /togglethinking.

• 📋 Multi-line Input:

  • End any line with \ to continue on the next line
  • Use /paste [text] for bulk paste mode (end with ---)

• 💾 Session Management:

  • /save <name> — Save session to ~/.chat_sessions/<name>.json
  • /load <name> — Load a saved session
  • /clear — Delete all saved sessions

• Python 3.9 or newer
• No pip installs required — uses only Python standard library

1. Clone the repository:

   git clone https://github.com/olumolu/consoleAI.git
   cd consoleAI

2. Configure API Keys:

   [!IMPORTANT] You MUST add your API keys. Open ai.py and locate the API_KEYS dictionary:

   API_KEYS: dict[str, str] = {
       "gemini":     "",   # https://aistudio.google.com/app/apikey
       "openrouter": "",   # https://openrouter.ai/keys
       "groq":       "",   # https://console.groq.com/keys
       "together":   "",   # https://api.together.ai/settings/api-keys
       "cerebras":   "",   # https://cloud.cerebras.ai/
       "novita":     "",   # https://novita.ai/
       "ollama":     "",   # Leave blank for local Ollama
   }

   Alternatively, set environment variables:

   export GEMINI_API_KEY="your-key-here"
   export OPENROUTER_API_KEY="your-key-here"
   # etc.

3. Run it:

   python ai.py <provider> [filter]...

python ai.py gemini
python ai.py groq llama
python ai.py openrouter claude
python ai.py together
python ai.py cerebras
python ai.py novita
python ai.py ollama

Use the optional [filter]... argument to narrow down model selection:

python ai.py openrouter 32b        # Shows only models with "32b" in the name
python ai.py gemini pro            # Shows only models with "pro" in the name

• End any line with \ to continue on the next line
• Use /paste for bulk paste mode (end with --- on its own line)

When enabled, the model can invoke local tools:

Note: Gemini also gets Google Search grounding when tools are enabled.

You can customize settings at the top of ai.py:

MAX_HISTORY_MESSAGES  = 20        # Messages to keep in context
MAX_MESSAGE_LENGTH    = 50_000    # Max chars per message
DEFAULT_TEMPERATURE   = 0.7
DEFAULT_MAX_TOKENS    = 3000
DEFAULT_TOP_P         = 0.9
SYSTEM_PROMPT         = "You are a helpful assistant running in a command-line interface."
ENABLE_THINKING_OUTPUT = True
MAX_TOOL_ITERATIONS   = 10

Runs on:

• macOS
• Linux
• Android (via Termux)
• Any system with Python 3.9+

The original bash-based version is still included for minimal environments.

• Multi-provider support (Gemini, OpenRouter, Groq, Together, Cerebras, Novita, Ollama)
• Dynamic model selection with filtering
• Conversation history + /history command
• Streaming responses
• Color-coded output
• Tool calling (Gemini)
• Session management (/save, /load, /clear)
• Image support (/upload, /image, /clearimage)
• Minimal dependencies: only bash, curl, bc, and jq

• bash
• curl
• bc
• jq

chmod +x ai.sh

./ai.sh gemini
./ai.sh groq llama
./ai.sh openrouter claude
./ai.sh together
./ai.sh cerebras
./ai.sh novita
./ai.sh ollama

Edit the API key section at the top of ai.sh:

GEMINI_API_KEY=""
OPENROUTER_API_KEY=""
GROQ_API_KEY=""
TOGETHER_API_KEY=""
CEREBRAS_API_KEY=""
NOVITA_API_KEY=""
OLLAMA_API_KEY=""