Popeye: Your AI Coding Agent, Right in the Terminal
Plan, understand, and build software without leaving your terminal. Popeye is a local-first AI coding agent that combines a focused command-line interface with the models you already trust.
Why Developers Choose Popeye
Popeye brings AI-assisted development into a fast, keyboard-focused terminal workflow. Inspect codebases, plan changes, edit files, and run shell commands under your complete local control.
Stay in Your Development Flow
Work where your code already lives. Popeye provides an interactive terminal interface for planning, coding, reviewing changes, and managing longer tasks without context-switching between browser tabs.
Choose the Right Model
Use Anthropic Claude, OpenAI, Azure OpenAI, or models served locally through Ollama. Switch models mid-work and adjust effort levels matching task complexity.
Keep Control of Project Access
Project tools are restricted to the current working directory. Plan mode focuses on read-only investigation, while Build mode enables file edits and execution when authorized.
Keep Sessions on Your Machine
Configuration and conversation sessions are stored in local SQLite. Return to earlier work anytime without depending on remote server storage or cloud accounts.
Extend the Agent with Skills
Add reusable Markdown-based skills for workflows such as brainstorming, TDD testing, security code reviews, debugging, or team-specific engineering standards. Skills load on demand to preserve context window.
Core Feature Capabilities
Explore Popeye's terminal capabilities built with Bun, OpenTUI, React, and SQLite.
Terminal AI Chat
Use a responsive OpenTUI and React interface built specifically for the command line. Responses stream as they are generated, Markdown remains readable, and tool activity stays visible without overwhelming the conversation.
Plan and Build Modes
Use PLAN mode to investigate a project with read-only tools. Switch to BUILD mode when Popeye should write files, apply edits, or execute shell commands. The separation makes the agent's current level of access clear.
Multiple AI Providers
Configure one or more supported providers and select models from the terminal. Popeye supports Anthropic, OpenAI, Azure OpenAI deployments, and Ollama models discovered from your local instance.
Adjustable Agent Effort
Choose from Low, Medium, High, X-High, and Max effort levels. Popeye maps the selected level to the capabilities of the active provider while preserving a consistent workflow when you switch models.
Reasoning and Thinking Views
Use extended thinking with supported Claude models, detailed reasoning summaries with supported OpenAI models, and thinking-capable Ollama models. Reasoning output is collapsible, so you can focus on the final result or inspect more detail when needed.
Sandboxed Project Tools
Popeye can read files, list directories, search by file pattern, search source text, write files, apply targeted edits, and run shell commands. Local tools are restricted to the current project directory.
Persistent Local Sessions
Conversations and messages are stored in a local SQLite database. Browse earlier sessions, reopen a task, and continue with its previous context from the /sessions command.
Context-Usage Meter
See how much of the active model's context window is in use. For local or custom models without a known fixed context window, Popeye displays the raw token count.
Structured Clarifying Questions
When an assumption could lead to unnecessary rework, Popeye can ask a focused multiple-choice or free-text question directly inside the terminal interface.
Task Tracking
Break larger requests into visible tasks, track their status, and keep implementation progress connected to the conversation. Task summaries help both the developer and the agent understand what is pending, active, completed, or blocked.
Themes & Keyboard Navigation
Choose a terminal theme, navigate dialogs from the keyboard, scroll through long conversations, and use a command menu to reach common actions quickly.
Reusable Skills
Add reusable Markdown-based skills for workflows such as brainstorming, testing, debugging, documentation, or project-specific engineering practices. Skills load when needed instead of filling every conversation with unnecessary instructions.
Supported AI Providers
Connect Anthropic Claude, OpenAI, or Azure OpenAI with your own credentials, or use Ollama to work with local models on your GPU/RAM.
Anthropic Claude
Cloud API IntegrationConnect your Anthropic API key and work with supported Claude Opus, Sonnet, and Haiku models. Supported reasoning models can use extended thinking with effort settings adapted to the provider.
Available Models:
Run this command in Popeye CLI or during interactive onboarding to configure:
$ popeye /config set anthropic --api-key="sk-ant-..."Essential Commands
Keyboard-first interaction through slash commands in the Popeye command menu.
| Command | What It Does | Copy |
|---|---|---|
| /new | Start a new conversation session. | |
| /agents | Switch between Plan and Build modes. | |
| /models | Choose from configured cloud and local models. | |
| /effort | Select the agent effort level for the current workflow. | |
| /sessions | Browse and reopen locally stored SQLite sessions. | |
| /config | Add or update provider settings and API credentials. | |
| /theme | Change the terminal color theme. | |
| /exit | Close Popeye CLI cleanly. |
Skills That Match the Way You Work
Popeye skills are reusable Markdown instruction packs. A skill can encode a testing discipline, debugging process, documentation standard, or team engineering guidance.
Personal Skills
Stored in your user directory and available across all your local projects. Perfect for individual productivity tools and custom prompts.
Project Skills
Live in .popeye/skills/ within your repository and committed to Git so the whole team shares identical guidelines.
Featured Pre-Built Skills:
# Brainstorming Skill When asked to design or solve a complex architectural question: 1. Deconstruct the requirements into core invariants. 2. Outline 2-3 distinct implementation tradeoffs. 3. Present clear pros, cons, and performance implications. 4. Recommend the optimal path before taking implementation action.
$ popeye --add-skill https://github.com/obra/superpowers.git skills/brainstormingFrequently Asked Questions
Everything you need to know about Popeye local AI coding agent architecture, model providers, and privacy.