Ollama Console Markdown Display

Chat with Ollama models in your terminal with beautifully rendered markdown output and clipboard-copyable code blocks. Powered by Bun v1.3.12's Bun.markdown.ansi() API.

Demo

Requirements

• macOS (ARM or Intel)
• Bun v1.3.12+
• Ollama installed (started automatically if not running)

Usage

One-liner mode

# Basic question
bun run index.ts "explain closures in JavaScript"

# Specify a model
bun run index.ts -m qwen3.5:4b "write fizzbuzz in python"

# Save response to a file
bun run index.ts -o response.md "create a REST API in Node.js"

# Enable thinking/reasoning mode
bun run index.ts -t "why is the sky blue"

# Combine flags: custom model + thinking + save output
bun run index.ts -m qwen3.5:27b -t -o answer.md "explain monads"

Interactive chat mode

# Start chat with default model
bun run index.ts

# Start chat with a specific model
bun run index.ts -m qwen3.5:27b

# Start chat with thinking mode enabled
bun run index.ts -t

# Start chat with auto-save to file
bun run index.ts -o session.md

# Combine: specific model + thinking + output
bun run index.ts -m qwen3.5:4b -t -o log.md

Read a markdown file

# Auto-detect .md file
bun run index.ts README.md

# Explicit flag
bun run index.ts -r notes.md

Show help

bun run index.ts -h

Options

Flag	Description
-m, --model <name>	Model to use (default: qwen3.5:latest)
-o, --output <file>	Save response markdown to a file
-r, --read <file>	Render a markdown file in the terminal
-t, --think	Enable thinking/reasoning mode
-h, --help	Show help

Chat commands

Command	Action
<n>	Copy code block #n to clipboard
all	Copy all code blocks to clipboard
save <file>	Save last response to a file
model <name>	Switch to a different model
models	List available Ollama models
think	Toggle thinking mode on/off
think on/off	Set thinking mode explicitly
exit / quit	End the chat

After each response with code blocks, a copy prompt appears where you can type a block number, all, or press enter to continue chatting.

Install as a global command

Bun's --compile flag produces unsigned ARM64 binaries that macOS kills on Apple Silicon. Use a shell wrapper instead:

# Make the wrapper script executable and link it into your PATH
chmod +x ollama-chat.sh
sudo ln -sf "$(pwd)/ollama-chat.sh" /usr/local/bin/ollama-chat

Now you can use it from anywhere:

ollama-chat "explain closures in JavaScript"
ollama-chat -m qwen3.5:27b -t "why is the sky blue"
ollama-chat README.md

Features

• Streaming — tokens appear in real-time, then get replaced with formatted markdown
• Rendered markdown — headings, lists, bold, italic, code blocks with language labels
• Read .md files — render any markdown file in the terminal with syntax coloring
• Code block clipboard — numbered code blocks with copy <n> to pbcopy
• Save output — save responses to markdown files with -o flag or save command
• Model switching — switch models mid-chat with model <name>, list with models
• Thinking mode — enable chain-of-thought reasoning with -t flag or think command
• Conversation memory — chat mode maintains full message history
• Automatic Ollama startup — detects if Ollama is running and starts it automatically if needed

Automatic Ollama Startup

The main function includes an enhanced Ollama startup process that handles server availability automatically:

1. Detection — on launch, the app checks if Ollama is already running by pinging http://localhost:11434/api/tags
2. Auto-start — if Ollama is not running, it spawns ollama serve in the background and waits up to 15 seconds for it to become ready
3. Error handling — if Ollama fails to start (e.g. not installed), an error message is displayed with a link to the installation page at https://ollama.ai
4. Graceful cleanup — signal handlers (SIGINT, SIGTERM, exit) ensure the spawned Ollama process is stopped when the app exits

No manual server management is required — just run the app and it takes care of the rest.

Note on compiled binaries

Bun's --compile flag currently produces unsigned ARM64 binaries that macOS kills on Apple Silicon. Use the shell wrapper install method above instead.