Skip to content

LatentGate — Local-First AI Compression for VS Code

Compress images, text, conversations, and documents locally via Ollama before sending to cloud LLMs. Save ~80% on token costs.

Features

Compress selected text — right-click or Ctrl+Shift+Alt+C
Compress images — right-click image files in explorer/editor
Compress documents — full file compression from command palette
Cost dashboard — track token savings per session
Ollama health check — verify your local setup
Selective decoding — skip API calls when content is unchanged
Embedding-based Similarity — more accurate scene detection

Requirements

Ollama running locally
At least one model pulled (e.g., ollama pull phi3:mini)

Extension Settings

Setting	Default	Description
`latentGate.ollamaBaseUrl`	`http://localhost:11434`	Ollama server URL
`latentGate.visionModel`	`llava:7b`	Vision model
`latentGate.textFastModel`	`phi3:mini`	Fast text model
`latentGate.textSmartModel`	`qwen2:7b`	Smart text model
`latentGate.embeddingModel`	`nomic-embed-text`	Embedding model
`latentGate.enableCaching`	`true`	Cache repeated images
`latentGate.selectiveDecoding`	`true`	Skip unchanged scenes

Commands

LatentGate: Compress Image
LatentGate: Compress Selected Text
LatentGate: Compress Current Document
LatentGate: Show Cost Dashboard
LatentGate: Show Session Stats
LatentGate: Open Settings
LatentGate: Check Ollama Health
LatentGate: Configure MCP Server