TokenDock

Your AI Context Control Center for GitHub Copilot and Cursor. Analyse prompts, trim noisy tabs, track session cost, and send better context — all locally, no data leaves your machine.

What You Get

📋 Context Tab

• Token status — See how many tokens your open tabs consume (green/yellow/red)
• Token-Optimized Prompt — Mentions a function like processPayment? TokenDock inlines the exact code so Copilot doesn't load whole files. Shows token savings vs full-file context
• Uncommitted Changes — Lists files with git changes so you can open them for Copilot to see your latest work
• Large Open Tabs — Big files eating your token budget. Close the ones you don't need
• Git History Insights — Scan your repo's commit history to find files that usually change together with what you're working on
• Mode Recommendation — Suggests Ask vs Agent vs Plan based on prompt complexity (Ask is 25× cheaper than Agent for simple questions)

✏️ Prompt Tips Tab

• 12+ rule-based checks — Catches vague language, missing #file: refs, filler phrases, multi-task prompts, no stopping condition, missing output format, and more
• Suggested Prompt — A rewritten version of your prompt that's clearer and more specific. Copy or send to chat with one click
• Local AI enhancement — Optionally use Qwen via Ollama for smarter prompt rewrites (rule-based works without it)

📊 Session Tab

• 💰 Session cost — Real $ spent in your current chat session
• ⚡ Response speed — Tracks how fast Copilot responds. If it's slowing down, start a new chat
• 🧠 Context window fill — How full your chat memory is. Over 80% = Copilot starts forgetting earlier messages
• ♻️ Cache reuse — % of tokens reused from cache. Higher = cheaper follow-ups. Stay in one chat to keep this high
• 🔧 Tool usage — Which Copilot tools (file reads, edits, terminal) are used most in your session
• 📜 Cross-session history — Aggregate stats across all your past Copilot chats in this workspace

How to Use

1. Click the 🤖 TokenDock icon in the Activity Bar (left sidebar)
2. Paste your prompt → click 🔍 Analyse (or Cmd+Shift+J / Ctrl+Shift+J)
3. Context tab — close noisy tabs, open missing files, review inline code extraction
4. Prompt Tips tab — expand any tip for details, copy the suggested prompt
5. Session tab — monitor cost, speed, and cache reuse

Works With

• VS Code — full support
• Cursor — works with fallback for Cursor's chat commands

Local AI Setup (Optional)

TokenDock is fully functional without Ollama — rule-based suggestions work out of the box. Enable Ollama only if you want Qwen 2.5 Coder to rewrite your prompts.

One-time install

brew install ollama
ollama pull qwen2.5-coder:7b

Turning Ollama ON

1. Cmd+Shift+P → TokenDock: Start Ollama (Qwen) (or open a terminal and run ollama serve yourself)
2. Settings (Cmd+,) → enable copilotAdvisor.enableLocalAiSuggestedPrompt
3. Suggested prompts now show 🤖 Qwen + rules

Turning Ollama OFF

1. Disable copilotAdvisor.enableLocalAiSuggestedPrompt (this alone stops all HTTP calls)
2. Cmd+Shift+P → TokenDock: Stop Ollama to kill the daemon (or run pkill -f "ollama serve" — Ollama has no built-in stop subcommand)
3. Suggested prompts fall back to 📋 Rule-based

Settings

Keyboard Shortcuts