Everything Gray does.

Installation

Install the Gray AI agent harness with one curl command on macOS, Linux or WSL, a digest-verified artifact on Windows 11, or build it from source.

Quickstart

From a fresh install to your first working AI agent conversation — connect a provider, pick a model, and start a session in a few steps.

Providers & models

Connect Gray to any OpenAI-compatible endpoint — OpenRouter, DeepSeek, OpenAI, Anthropic, Cerebras — and switch models mid-session.

Sessions & resume

Sessions are JSONL transcripts on disk with parent-id branching, so you can resume any AI agent conversation at exactly the point it stopped.

Context window & auto-compact

How Gray resolves an AI agent's context window, and how auto-compact summarizes older turns before you hit the limit and lose the thread.

Bash tools

One tool by default: bash. Read, search, edit and run commands through a single surface, with no tool sprawl and no per-tool approval prompts.

Agent skills

Skills are markdown procedures your AI agent discovers and runs on demand — a place for anything you would otherwise re-explain every session.

Gateway daemon

The gateway daemon runs Gray in the background: cron jobs, scheduled prompts and connections keep firing with no REPL window open.

Cron jobs

Schedule recurring prompts for your AI agent and let the gateway run them unattended on a cron schedule.

Slash commands

Every slash command in the Gray REPL — connect, model, context, resume, compact, skills, plugins, gateway and more.

Environment variables

Every environment variable the Gray AI agent reads at startup, from the context window to the API key, registry URL and log level.

Safety

What the Gray AI agent does and does not protect you from: it runs shell commands with your privileges and has no command guard.

Architecture

How the Gray crates fit together — the agent loop, provider streaming, the bash tool surface, plugins and session storage.