AgentiLoopGo 0.0.1-pre

⚠️ Pre-release — 0.0.1

⚠️ Pre-release

This is the first pre-release of AgentiLoopGo, built from main and tagged v0.0.1. AgentiLoopGo is the Go version of AgentiLoop CLI: a cross-platform AI coding agent for the terminal, in the spirit of Claude Code. It has the same features and options as the Rust version and uses the same settings, session and MCP files. It supports Anthropic, OpenAI-compatible servers (OpenAI, Ollama, LM Studio, …) and oMLX, and includes a full-screen TUI, saved sessions, and an MCP client ported from Agent!'s AgentMCP. The Release workflow builds for macOS, Linux and Windows. The macOS binaries are signed with Developer ID and notarized. Everything below is new in this release, so expect rough edges.

✨ New

• Agent loop: streaming (SSE) responses, tool calls, a permission prompt before anything is changed, and context compaction at --compact-at tokens or on demand with /compact (aef2ccf).
• Built-in tools: read_file, write_file, edit_file, list_dir, bash (aef2ccf).
• Providers: Anthropic (API key or Claude Code OAuth token, detected automatically), OpenAI-compatible (OpenAI, Ollama, LM Studio, Groq, OpenRouter, …), and oMLX (reads its port and API key from ~/.omlx/settings.json) (d133148).
• MCP client: stdio, Streamable HTTP and legacy HTTP+SSE servers from ~/.agentiloop/mcp.json / .mcp.json. Their tools appear as mcp_<server>_<tool>, and /mcp lists them. Includes an example server for all three transports (7b73808).
• TUI (--tui): transcript with Markdown and syntax highlighting, clickable links, a permission modal, and an animated ✻ Thinking... 12s indicator (2c991e7).
• Sessions and remembered launch: every turn is saved (-c, -r <id>, /sessions, /resume). A bare agentiloop reuses your last provider, model, TUI setting and session, and shows the earlier conversation again. --new and --no-tui override that (2c991e7).
• Slash commands: /model (live model list, pick by number), /clear, /compact, /sessions, /resume, /mcp, /help (2c991e7).
• CI and releases: tests run on macOS, Linux and Windows. Release builds cover 5 platforms, and macOS signing and notarization are required (41a4dd4, 6b69577).

🐛 Fixes

• bash tool: a timeout now kills the whole process tree (process group on Unix, taskkill /T on Windows), so child processes don't outlive the command (4b312a0).

📦 Downloads

Platform	File
macOS Apple Silicon	agentiloop-macos-arm64.tar.gz (signed + notarized)
macOS Intel	agentiloop-macos-x86_64.tar.gz (signed + notarized)
Linux x86_64	agentiloop-linux-x86_64.tar.gz (static, any distro)
Linux ARM64	agentiloop-linux-arm64.tar.gz (static, any distro)
Windows x86_64	agentiloop-windows-x86_64.zip

Unpack the download and put agentiloop (agentiloop.exe on Windows) on your PATH. You can also install from source with go install github.com/AgentiLoop/AgentiLoopGo/cmd/agentiloop@v0.0.1.

🧪 What to test

• Download the build for your platform, run agentiloop --version (expect 0.0.1), and confirm macOS runs it without a Gatekeeper warning.
• Set ANTHROPIC_API_KEY or OPENAI_API_KEY, or start oMLX, then run agentiloop --tui and give it a small task in a scratch repo.
• Approve and deny a few permission prompts (y / n / a / Esc).
• Quit and run a bare agentiloop: it should reopen with the same provider, model and TUI, and show the previous conversation.
• Switch between the Go and Rust builds in the same folder: both read ~/.agentiloop/ and should pick up each other's sessions.
• Add an MCP server to ~/.agentiloop/mcp.json and check /mcp.
• Windows SmartScreen may warn on first run because the Windows binary is not code-signed yet.
• Report problems with your OS, provider/model and the command you ran. Do not include API keys.

🔢 Version

• Version set to 0.0.1 (2c991e7).

Full Changelog: https://github.com/AgentiLoop/AgentiLoopGo/commits/v0.0.1