gemini-cli - 💡(How to fix) Fix Direct Integration of Gemini CLI as a Model Provider for Antigravity IDE. [1 participants]

What would you like to be added?

This architectural shift—moving from a "split-brain" setup (IDE vs. CLI) to a unified Agentic Orchestration—is exactly where high-end vibe coding is heading. By using Antigravity (AG) as the orchestrator and the Gemini CLI as the coder/subagent, you solve the context-drift problem entirely.

Here is the updated feature request, specifically focused on Orchestration & Provider Unification.

What would you like to be added? Direct Provider Integration for Gemini CLI to enable Unified Agentic Orchestration.

I am requesting that Gemini CLI be added as a first-class model provider within the Antigravity IDE. This is not just for chat, but to enable a Leader-Follower agentic workflow where:

Antigravity acts as the Orchestrator/Planner: Using its 1M+ context and Tree-sitter indexing to map out the project topology and high-level strategy.

Gemini CLI acts as the Coder/Subagent: Executing the specific implementation loops, leveraging its native subagent capabilities (v0.38.1+), and respecting local toolchains.

Unified Context Session: Both entities must share a single session state so that changes made by the CLI coder are immediately reflected in the AG planner’s context window without manual re-indexing or session drift.

Why is this needed?

Currently, running Antigravity and the Gemini CLI separately causes Session Drift:Context Fragmentation: AG might plan a refactor based on state $A$, while the CLI (with its own subagents and planning mode) is already at state $B$. This leads to hallucinated file paths and broken dependency chains.Redundant Subagent Logic: Gemini CLI recently introduced powerful Subagents (@codebase_investigator, @generalist). If AG cannot call the CLI as a provider, I am forced to manually sync these subagent findings back into the IDE.Configuration Silos: My model steering and "vibes" are defined in the CLI's logic. By using the CLI as the provider, Antigravity gains access to the exact model behaviors I've tuned in the terminal.

Additional context

The implementation should leverage the existing Gemini CLI Companion logic but in reverse: instead of the CLI "looking" into the IDE, the IDE should "drive" the CLI.Proposed Integration Logic:Subagent Delegation: In the AG chat, I should be able to trigger CLI subagents directly (e.g., /cli @codebase_investigator "Find the bottleneck in our DuckDB queries").Agentic Handover: AG creates the plan => Passes the plan to Gemini CLI => CLI executes via its internal Planning Mode => Results stream back into the AG terminal and editor.Local-First Sync: Ensure that the .agents/rules/ directory is the "Source of Truth" for both the AG orchestrator and the CLI coder.