Overview

Jido Code is an agentic coding assistant that runs entirely in your terminal, built on the Jido autonomous agent framework. It combines LLM-powered chat with a rich set of sandboxed tools — file operations, search, shell commands, web fetching, and Livebook editing — all wrapped in an Elm Architecture TUI with real-time streaming, Chain-of-Thought reasoning, and multi-session support.

Unlike hosted coding assistants, Jido Code is self-hosted and provider-agnostic. It supports Anthropic, OpenAI, and any provider registered through JidoAI, with runtime switching between models. Its multi-layered security model ensures the agent can only operate within your project’s boundaries, making it safe for autonomous tool use.

Purpose

Jido Code is the Jido ecosystem’s coding agent — an interactive, terminal-based AI assistant that demonstrates the full capabilities of the Jido agent framework in a real-world developer tool. It serves as both a practical coding assistant and a reference implementation.

Major Components

TUI Layer

Elm Architecture terminal interface built on TermUI with conversation view, session tabs, sidebar, markdown rendering with syntax highlighting, and keyboard shortcuts.

Agent Layer

LLMAgent wrapping JidoAI’s agent system for chat interactions, streaming, tool dispatch, and conversation history. TaskAgent for isolated sub-task execution with independent LLM context.

Tool System

16 tools with ETS-backed registry, executor with timeout enforcement, and Lua sandbox for secure script execution. Covers file read/write/edit, directory listing, search, shell, Livebook, web, todo, and task operations.

Security Layer

Multi-layered enforcement: path validation against project boundaries, shell command allowlists, protected settings files, web domain restrictions, and Lua sandbox with restricted stdlib.

Session System

Multi-session support with DynamicSupervisor lifecycle management, per-session state and persistence, and isolated security boundaries.