Gemini Relay • Antigravityv1.2.0

Gemini RelayGoogle Gemini 3.8 & Antigravity for AI Agents

Unleash Google's Gemini 3.8 Flash, Gemini 3.1 Pro, reasoning depth, and Imagen multimodal generation directly inside Claude Code, Cursor, and autonomous agent workflows.

Quick Start →

🤖 Agent Reference Guide

GitHub Repo ⭐

⚡

Gemini 3.8 Flash by Default

Instantaneous response latency coupled with deep multi-step thinking tokens via gemini-3.8-flash-high.

🧠

Dynamic Reasoning Depth

Dial thinking token depth up or down on demand with effort controls ('low', 'medium', 'high') per query.

🏗️

Architectural Blueprints (gemini-plan)

Non-destructive phased implementation roadmaps and dependency risk assessments using Gemini's dedicated plan mode.

🎨

Multimodal Visual Creation (gemini-image)

Generate application assets and illustrations from text descriptions with aspect ratio control and automatic project export.

📐

Guaranteed JSON Schemas

Enforce valid machine-readable JSON matching any user-provided schema for zero-friction agent parsing.

📂

Zero-Overhead Context Inlining

Offload 1M-2M tokens of code and logs via @file syntax, keeping your primary agent's context clean and hyper-focused.

🎮 Interactive Agent Console

Experience how Claude Code interacts with gemini-relay in real-time across architectural planning, large context inlining, visual generation, and structured outputs.

🤖 Claude Code <—> Gemini Relay Console

agy v1.1.25 • Gemini 3.8

Claude Code Querygemini-plan(task, effort: "high")

"Use gemini-plan to architect a low-latency websocket streaming server with automatic token refresh and backpressure."

Model:gemini-3.8-flash-high

Thinking:4,180 tokens

Duration:3.4s

Status:SUCCESS 200

### 📋 Architectural Blueprint: Resilient WebSocket Server
**Mode:** --mode plan | **Reasoning Effort:** High (4,180 thinking tokens)

#### Phase 1: Core Stream Engine & Backpressure
- Implement stream.Writable with highWaterMark: 64KB.
- Buffer overflow strategy: Pause socket frame ingestion when client buffer exceeds 80% threshold.

#### Phase 2: Autonomous Token Rotation
- Background lease manager checking JWT expiry at T-60s.
- Non-blocking handshake refresh without closing established TCP sockets.

#### Phase 3: Failure Modes & Recovery
- Exponential backoff with full jitter: min(30s, 100ms * 2^attempt + rand(0, 50ms)).

🧠 Reasoning Effort Depth Engine

Explore the thinking token budgets available across Gemini 3.8 Flash and Gemini 3.1 Pro.

Control thinking token depth per tool call using effort: "low" | "medium" | "high".

Thinking DepthDeep Algorithmic Thinking

Typical Latency~4.0s - 8.0s

Recommended Modelgemini-3.8-flash-high

🎯 Ideal Workflows:

Security vulnerability audits, race conditions, distributed architecture design, mathematical algorithms, and comprehensive system planning (gemini-plan).

MCP Tool Call Configurationeffort: "high"

await useGeminiPlan({
  task: "Design distributed lock mechanism with Redis & auto-renew",
  context: "High-throughput financial ledger API with strict consistency",
  effort: "high"
});

🌟 Supported Gemini 3.8 & 3.1 Model Matrix

DEFAULT CHOICE

gemini-3.8-flash-high

Flagship Fast & Deep

The default engine for Gemini Relay. Delivers state-of-the-art multimodal speed paired with multi-step reasoning depth.

Reasoning Depth:High (Deep Thinking)

Context Window:1M+ Tokens

Execution Speed:Ultra-Fast (~2.5s)

DefaultCode ReviewPlanningImage Gen

MAX REASONING

gemini-3.1-pro-high

Deep Architectural Thinker

Top-tier complex problem solving, formal mathematical algorithms, concurrency analysis, and mission-critical architecture.

Reasoning Depth:Maximum Flagship

Context Window:2M+ Tokens

Execution Speed:Deliberate (~5-8s)

Security AuditsDistributed SystemsProofs

INSTANT THROUGHPUT

gemini-3.8-flash-low

Zero-Latency Engine

Optimized for instant parsing, commit summaries, AST linting, and high-frequency queries without thinking delay.

Reasoning Depth:Low / Instant

Context Window:1M+ Tokens

Execution Speed:Near-Instant (<1s)

High QPSSummariesFormatting

⚡ One-Line Setup for Claude Code

Add the MCP server to your local Claude Code instance in seconds:

bash

claude mcp add gemini-relay -- npx -y gemini-relay

Verify your setup at any time by asking Claude:

"Run gemini-doctor to check the status of the MCP server and active backend"