The agent engine is the shared core layer that runs underneath both Antigravity CLI and the Antigravity 2.0 GUI. Rather than maintaining two separate reasoning stacks, Antigravity ships a single engine that both surfaces call into — so every capability improvement, model upgrade, or bug fix you receive in one interface automatically benefits the other.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/google-antigravity/antigravity-cli/llms.txt
Use this file to discover all available pages before exploring further.
What the agent engine provides
The engine exposes four core capabilities to every interface built on top of it:- Multi-step reasoning — The agent breaks complex tasks into smaller steps, working through them sequentially and asking for your approval before applying changes.
- Multi-file editing — A single agent turn can read, propose edits across, and write to multiple files in your codebase. Diffs are presented before any change is committed.
- Tool calling — The agent can invoke terminal commands, search your codebase, and interact with external tools during a session. All tool calls are subject to your configured tool permissions.
- Persistent history — Sessions and their full context are stored so you can resume exactly where you left off. See Session management for details.
Why both interfaces share the engine
Antigravity CLI and Antigravity 2.0 run on the same core agent engine. Improvements to the engine — new models, smarter reasoning, faster tool execution — propagate to both surfaces automatically without any action required on your part.
- Consistent behavior — A coding task you start in the terminal and export to the GUI continues with the same agent logic. You don’t get a different quality of reasoning depending on which surface you use.
- Synchronized settings — Preferences, permissions, and model configuration sync bidirectionally between the CLI and the GUI. Configure once; both surfaces reflect the change.
- No duplicate maintenance — Google engineers maintain one engine instead of two, which means more development time goes toward capability improvements rather than keeping two codebases in parity.
How improvements propagate
When Google ships an engine update — for example, an improvement to how the agent handles long-context files or a new model version — you receive it automatically the next time you update the CLI. Because the GUI draws from the same engine, the update reaches both surfaces in the same release cycle. This also means that feedback you provide in either surface (flagging a bad edit, approving a tool call, adjusting a permission) can inform the same underlying system.Explore related topics
Session management
Learn how sessions store history, fold long inputs, and export to the GUI.
Tool permissions
Understand how the agent requests approval and how the sandbox works.
Authentication
Sign in locally or over SSH, including enterprise GCP access.
Command reference
Browse all slash commands for sessions, account, settings, and more.