echoVic/orca-agent

What it solves

Orca is a terminal-based coding agent designed to automate software development tasks. It eliminates the manual loop of editing files, running tests, and verifying results by allowing an AI agent to perform these actions directly in the terminal, reducing the friction of switching between an editor and a command line.

How it works

Built in Rust, Orca leverages DeepSeek's reasoning and tool-use capabilities to interact with the local file system and shell. It can read, search, and edit code, execute shell commands, and verify the outcome of its actions using specified commands. The agent operates in several modes: suggest (for guidance), auto-edit (for sandboxed edits), full-auto (for autonomous operation), and plan (for read-only planning).

Who it’s for

Developers who want an AI-powered coding assistant that lives in their terminal and can autonomously handle tasks like fixing bugs, running tests, and others, while maintaining control over the security and permission levels of the process.

Highlights

• DeepSeek-native: Optimized for DeepSeek's reasoning and tool-use semantics, including support for reasoning effort levels.
• Flexible Execution: Supports both an interactive TUI (Terminal User Interface) and a headless mode (orca exec) for scripts and CI/CD pipelines.
• Autonomous Workflows: Capable of spawning synchronous and asynchronous subagents and executing JavaScript workflows for complex, long-running tasks.
• Coded Memory: Learns and retrieves durable project-specific facts to maintain context across sessions.
• Multimodal Support: Supports image inputs (JPEG, PNG, GIF, WebP) for visual analysis of UI or errors.
• Robust Persistence: Conversations can be resumed, forked, or archived, with a digest-verified checkpoint system for reliability.
• Extensibility: Integrates with MCP (Model Context Protocol) tools and resources, and supports custom plugins and skills.