gptme — AI Agent in Your Terminal

What it is

gptme is an open-source personal AI assistant that operates entirely inside your terminal. It connects to OpenAI, Anthropic, or local models and can write code, run shell commands, browse the web, and manage files. Unlike cloud-based AI assistants, gptme keeps everything local and transparent.

It targets developers who want an AI pair programmer without leaving the command line. If you prefer terminals over GUIs and want full control over what the AI executes, gptme fits that workflow.

How it saves time or tokens

gptme reduces context-switching by keeping the AI conversation, code editing, and command execution in a single terminal session. Instead of copying code between a chat window and your editor, gptme writes files and runs commands directly. The tool logs all conversations to a local SQLite database, so you can resume sessions and avoid re-explaining context. Token usage stays efficient because gptme streams responses and only sends relevant context.

How to use

1. Install gptme via pip and set your API key:

pip install gptme
export OPENAI_API_KEY='sk-...'
# Or: export ANTHROPIC_API_KEY='sk-ant-...'

1. Start a conversation with a task:

gptme 'Create a Python script that scrapes HN top stories'

1. Resume a previous conversation or use specific features:

# Resume last conversation
gptme --resume

# Use a specific model
gptme --model anthropic/claude-sonnet 'Refactor this module'

# Non-interactive mode for scripting
gptme --non-interactive 'Write tests for app.py'

Example

# Start gptme and ask it to build a web scraper
gptme 'Build a Python web scraper that extracts titles from news.ycombinator.com'

# gptme will:
# 1. Write a Python script with requests + BeautifulSoup
# 2. Save it to a file
# 3. Run it and show you the output
# 4. Fix any errors automatically

# You can then follow up:
# > 'Add error handling and save results to CSV'

Related on TokRepo

• AI coding tools -- Browse other AI-powered development tools
• CLI and automation tools -- More command-line AI utilities

Common pitfalls

• gptme executes commands on your real system by default. Use the sandbox mode or Docker integration for untrusted tasks to avoid accidental file modifications.
• API key management matters. If you switch between providers frequently, use environment variables or gptme's built-in key management rather than hardcoding keys.
• Large codebases can exhaust context windows quickly. Break tasks into smaller, focused requests rather than asking gptme to refactor an entire project at once.