Back to Projects
Case Study
Axiom
Terminal Coding Agent with Multi-Provider LLM Support
TypeScriptNode.jsReactLLM APIsMCP Protocol
Overview
Axiom is a powerful terminal-based coding agent inspired by Claude Code CLI. It features a premium TUI with streaming responses, multi-provider LLM support (OpenCode, Anthropic, OpenAI, Google, Groq, xAI, Cerebras), tool execution with security validation, vim mode, collapsible thinking panels, and session management with branching support. Built as a monorepo with separate packages for AI, agent-core, TUI, and the coding agent itself.
Features
- Premium TUI with Claude Code CLI-style interface and 60fps rendering
- Multi-provider support — OpenCode, Anthropic, OpenAI, Google, Groq, xAI, Cerebras
- Tool execution — read, write, bash, edit, grep, find, ls, mkdir with security validation
- Security-first design — dangerous command detection, path validation, read-only checks
- Full vim keybindings (i, Esc, h/j/k/l, w, b, dd, dw)
- Collapsible thinking panel — Tab key to toggle reasoning display
- Session management with JSONL persistence and branching support
Challenges
- Building a 60fps terminal UI framework with dirty region optimization
- Implementing consistent tool execution across multiple LLM providers with different API characteristics
- Security validation — accurately detecting dangerous commands without blocking legitimate ones
- Session branching — managing complex conversation trees with JSONL persistence
Learnings
- Terminal UI rendering techniques including frame-based rendering and dirty region optimization
- Multi-provider LLM abstraction — handling different streaming formats, token limits, and error modes
- Security-first tool execution design patterns
- Monorepo architecture with shared packages and build orchestration