EverRun ⚡

A VS Code extension built for one goal: Never let your coding flow stop.

Features

Feature	Description
Prompt Compression	Automatically trims and compresses conversation history before sending to Claude
Smart Context Injection	Detects relevant open files and injects only the useful parts
Conversation Summarisation	Collapses old messages into dense summaries instead of dropping them
Streaming Responses	Typewriter-effect streaming for immediate feedback
Save & Resume Conversations	Persist any conversation to disk and reload it later
Task Notes	Attach free-text notes to a session so you remember where you left off
Settings Panel	Full configuration UI inside the extension

Quick Start

1. Install

# From VSIX (after packaging):
code --install-extension claude-token-optimizer-0.1.0.vsix

Or press F5 in VS Code with this folder open to launch the Extension Development Host.

2. Set your API Key

Open the Command Palette → EverRun: EverRun Settings
Or click the ⚡ Claude status bar item → gear icon in the panel

You need an Anthropic API key.

3. Send your first prompt

Select code in any editor
Press Ctrl+Shift+C (Cmd+Shift+C on macOS), or right-click → Optimize Prompt
The Chat Panel opens with your selection pre-filled
Type your question and hit Send

Commands

Command	Keybinding	Description
`EverRun: Optimize Prompt`	`Ctrl+Shift+C`	Open chat panel (pre-fills editor selection)
`EverRun: Open EverRun Chat Panel`	—	Open the chat panel directly
`EverRun: Save Conversation`	—	Save current history to disk
`EverRun: Load Conversation`	—	Pick and restore a saved conversation
`EverRun: Clear Conversation History`	—	Wipe in-memory history
`EverRun: EverRun Settings`	—	Open the VS Code settings page

Configuration

All settings are under the claudeTokenOptimizer namespace:

Setting	Default	Description
`apiKey`	`""`	Your Anthropic API key
`model`	`claude-3-5-sonnet-20241022`	Model to use
`maxTokens`	`4096`	Max response tokens
`maxHistoryMessages`	`10`	History window size
`enableContextCompression`	`true`	Toggle compression
`enableFileContext`	`true`	Attach open files
`summarizeOldMessages`	`true`	Summarise vs. drop old messages
`maxContextFileLines`	`100`	Lines per context file
`systemPrompt`	see default	System prompt sent with every request
`saveConversationsPath`	`""`	Override conversations save directory

How Compression Works

User Prompt + Conversation History
         │
         ▼
  ┌─────────────────────────────┐
  │  1. Trim history window     │  → keep only the N newest messages
  │  2. Summarise older half    │  → ask Claude to condense, not drop
  │  3. Attach relevant files   │  → keyword-scored, capped at N lines
  │  4. Build final message     │  → prompt + code_context block
  └─────────────────────────────┘
         │
         ▼
    Claude API (streaming)
         │
         ▼
    Typewriter response in VS Code panel

Token savings are shown in the stats bar after every response.

Development

# Install dependencies
npm install

# Compile TypeScript
npm run compile

# Watch mode (recompiles on save)
npm run watch

# Run in development (press F5 in VS Code)
# Package as VSIX
npm run package

Project Structure

src/
├── extension.ts                 # Entry point, command registration
├── commands/
│   └── optimizePrompt.ts        # "Optimize Prompt" command logic
├── services/
│   ├── claudeService.ts         # Anthropic SDK wrapper + token estimation
│   ├── contextCompressor.ts     # Compression pipeline
│   └── conversationManager.ts  # Save / load / history management
├── panels/
│   └── chatPanel.ts             # WebView panel (HTML + message bridge)
└── config/
    └── settings.ts              # Typed settings accessors

Privacy

Your API key is stored in VS Code's global settings (not transmitted anywhere except Anthropic).
Conversations saved to disk stay local; no telemetry is collected.

License

MIT

EverRun

hmadhsan