Terminal-native · Built in Go · GPL-3.0

The AI coding agent
that lives in your terminal.

Goa brings full LLM transparency, multi-agent collaboration, and a rich ANSI TUI to any OpenAI-compatible model — served to your browser, attachable from any terminal, with no opaque black boxes.

Get Started View on GitHub
$ go install github.com/pijalu/goa/cmd/goa@latest Copy
Go 1.25+ GPL-3.0-or-later macOS Linux Windows
What is Goa

Complete control over your AI workflow

Goa is a terminal-native AI coding agent. It gives you a full-featured chat interface in your terminal, powered by any OpenAI-compatible LLM. Every thought, every tool call, every result is visible — no hidden context, no surprises.

Designed for transparency

  • › Full LLM visibility — reasoning streams, tool I/O, and tokens are always inspectable.
  • › Provider-agnostic — OpenAI, llama.cpp, LM Studio, Ollama, any compatible endpoint.
  • › Composable tools — read, write, edit, search, bash, SSH, and more, all auditable.
  • › Open source — GPLv3, no telemetry, your data stays local.
goa — coder
Animated demo of Goa running in a terminal: streaming assistant output, tool calls, and diffs.
Features

Everything you need, nothing hidden

A complete agentic toolkit built on small, composable primitives — the way Go intended.

Multi-Provider

Works with any OpenAI-compatible endpoint: OpenAI, llama.cpp, LM Studio, Ollama, and more.

Tool System

Read, write, edit, search, bash, SSH, background execution, git utilities — all transparent.

Config Cascade

Embedded to home to project to local to environment to CLI overrides — predictable precedence.

Agent Profiles

Built-in coder, planner, and reviewer profiles — compose your own via inheritance.

Skills

Reusable prompt templates — inline system-prompt injection or sub-agent execution mode.

Multi-Agent

Pair (planner to coder) and reviewer (coder to reviewer) collaboration pipelines.

JS Plugins

Extend Goa with JavaScript via the Goja runtime — custom tools, commands, and UI.

Rich TUI

Differential-rendering ANSI UI: chat, thinking stream, tool ledger, side panels, modals.

Web UI

goa server serves the same session as a web page — HTML blocks + live cell band, token auth, read-only mode.

TUI/WEB Interop

One session, every surface: browsers and attached terminals (goa attach) watch and drive the identical conversation.

Multi-Project Server

--server-projects-root fronts one full session per project — path ACL, idle reaping, process isolation.

Session Tree

Visualize, clone and fork session history trees — every branch of a conversation, kept.

Swarm & Pair Mode

Multi-agent swarm with fan-out/fan-in, plus the planner + coder pair pipeline.

Thinking Levels

Reasoning effort per provider: off, low, medium, high — and thinking-block visibility.

Context Compression

Compress conversation history to fit the context window without losing the thread.

Steering

Guide agent behavior mid-session — queued input reaches the agent between steps.

History Persistence

Input history saved across sessions and restored into the editor.

Model/Provider Mgmt

Delete models and providers via /config:remove or backspace in the pickers.

Priority Search

Search results ordered by file type (source first) with match counts.

Configurable Spinner

50+ spinner styles via /config or tui.spinner — the title bar animates too.

Interactive Clarify

The agent asks structured clarifying questions answered on the input line, not guessed at.

Session Persistence

Full JSONL session history with /save and /restore — resume any conversation.

Diagnostic Export

Self-contained ZIP bundle via /export with events, logs, config, and issue description.

SOLID by design

Small primitives and factories over fat per-type APIs — the codebase is built to extend.

Orchestrator

Hub, fanout, and pipeline topologies with bounded agent pool, goal binding, and headless resume.

Workflows

Multi-stage, multi-agent pipelines with team collaboration and AgentBus messaging.

Goals & Plans

Autonomous goals with lifecycle and budgets, plus structured work plans with review pager.

Companion

Sub-agent code review — automatic diff review on every change, catching issues early.

Execution Modes

YOLO (auto-approve), confirm (pause before each tool), review (queue writes for batch approval).

Loop Detection

Five heuristics detect stuck agents: identical calls, token budget, timeout, activity, error rate.

Git Worktree Isolation

Sandboxed agent filesystem via git worktree — changes stay isolated until approved.

Architecture

A layered engine you can trace

From keystroke to LLM token, every byte path crosses a clearly separated layer.

GOA PROCESS TUI Layer ANSI engine · differential rendering · CSI 2026 synced output Header · ChatViewport · StatusBar · Editor · Footer Goa Core Engine orchestration · config cascade · session state CommandRouter · AgentManager · ConfigLoader · LoopDetector · ExecutionController · SessionStore Agent SDK agent lifecycle, event loop, and inter-agent bus Agent / Session SkillRunner · AgentBus Tool Registry composable, SOLID primitives read · write · edit · search · bash ssh · python · bg_exec · webfetch · agent Provider Layer any OpenAI-compatible endpoint OpenAIProvider · OpenAI · llama.cpp · LM Studio · Ollama
Quick Start

Up and running in three steps

No accounts, no API keys to us — just Go, your model endpoint, and your terminal.

  1. Build from source
    Clone the repository and compile the binary.
    $git clone https://github.com/pijalu/goa$cd goa && make build
  2. Run the interactive wizard
    On first run, Goa walks you through configuring a provider, selecting a profile, and choosing an execution mode.
    $./goa
  3. Start coding
    Ask questions, request refactors, review diffs — all from your terminal, with full visibility into every step.
Download

Pre-built binaries for every release

Grab the latest build from the GitHub Releases page for your platform.

linux / amd64 linux / arm64 macOS / amd64 macOS / arm64 windows / amd64
Documentation

Read the manual

Every layer of Goa is documented, from high-level architecture down to individual tools.

ArchitectureLayered engine design Setup GuideInstall and configure ConfigurationThe config cascade CommandsSlash commands reference ToolsThe tool registry SkillsReusable prompt templates TUITerminal interface ProfilesAgent personas
Browse all documentation