TokenIQ - AI Usage Efficiency Tracker
Know if your AI is actually helping you ship, or just burning tokens.
TokenIQ tracks every AI call you make inside VS Code and measures whether it produced real outcomes - commits, file saves, shipped code. No cloud. No account. No prompts stored.
Why TokenIQ?
You open Copilot Chat. You ask the same question five different ways. Two hours later you've made one commit. Was that efficient?
Most developers have no idea how much they're using AI, which models they're burning through, or whether the AI sessions that feel productive actually are. TokenIQ gives you the numbers.
What it tracks
- Every AI call - model used, response time, cost tier (Premium / Standard / Fast)
- Every dev outcome - git commits and file saves, weighted by type
- Efficiency ratio - AI calls ÷ dev actions. Lower means your AI usage is driving real work
- Real token counts - exact tokens in/out when using the built-in proxy for direct API calls
- Session history - see which work sessions were productive and which were rabbit holes
The dashboard
Click the TokenIQ icon in the VS Code activity bar to open the dashboard.
Active session shows what's happening right now:
- How many AI calls you've made
- Which models and their cost tiers
- Your current efficiency ratio
- Token counts when available via proxy
Efficiency ratio is the core metric:
| Ratio |
What it means |
| 0–2× |
AI is directly driving shipped work |
| 2–5× |
Some exploration, still productive |
| 5–10× |
Lots of AI, few outcomes |
| 10×+ |
Possible rabbit hole - step back |
Quick start
- Install TokenIQ from the VS Code Marketplace
- Open a project and start using GitHub Copilot Chat as normal
- TokenIQ starts capturing automatically - no configuration needed
- Click the bar chart icon in the activity bar to see your dashboard
No API keys, no sign-up, no configuration required for Copilot users.
Real token counts (optional)
Copilot Chat does not expose token counts. If you also call OpenAI, Anthropic, or Ollama APIs directly from your own scripts, TokenIQ's built-in proxy captures exact token counts.
Enable it in settings:
"tokeniq.proxy.enabled": true
Reload VS Code. TokenIQ automatically sets OPENAI_BASE_URL and ANTHROPIC_BASE_URL in every new terminal - no manual environment variables needed. Your scripts route through http://localhost:8888 on the way to the real API.
Privacy: the proxy reads only token counts and model names. It never stores prompts or AI responses.
Privacy - the short version
TokenIQ stores everything locally in a SQLite database on your machine. Nothing is sent anywhere.
| Stored |
Never stored |
| Model names |
Prompt text |
| Token counts |
AI response text |
| Response latency |
File contents or code |
| Git branch names |
Full file paths |
File extensions (e.g. .ts) |
Anything personal |
Configurable redaction for sensitive projects:
"tokeniq.privacy.redactWorkspaceNames": true,
"tokeniq.privacy.redactBranchNames": true,
"tokeniq.privacy.excludedWorkspaces": ["client-acme", "confidential"]
Supported AI providers
| Provider |
Capture method |
Token counts |
| GitHub Copilot Chat |
Automatic |
Not available (Copilot limitation) |
| OpenAI API (direct calls) |
Proxy |
✅ Exact |
| Anthropic API (direct calls) |
Proxy |
✅ Exact |
| Ollama (local models) |
Proxy |
✅ Exact |
| Continue.dev |
Automatic |
When exposed by Continue |
Settings reference
| Setting |
Default |
Description |
tokeniq.proxy.enabled |
false |
Start local proxy for direct API token tracking |
tokeniq.proxy.port |
8888 |
Proxy port (auto-increments if in use) |
tokeniq.dataRetentionDays |
90 |
Delete sessions older than N days |
tokeniq.privacy.redactWorkspaceNames |
false |
Store [workspace] instead of real name |
tokeniq.privacy.redactBranchNames |
false |
Store [branch] instead of real name |
tokeniq.privacy.excludedWorkspaces |
[] |
Workspace names to exclude from tracking |
tokeniq.privacy.customPatterns |
[] |
Extra regex patterns to redact from metadata |
tokeniq.maintenance.vacuumIntervalDays |
7 |
How often to run SQLite VACUUM |
tokeniq.maintenance.warnDbSizeMb |
50 |
Warn when local DB exceeds N MB |
Diagnostics
If capture stops working after a Copilot update:
Ctrl+Shift+P → TokenIQ: Run Diagnostics
Shows which provider levels are active, proxy status, and step-by-step fix instructions.
Requirements
- VS Code 1.96 or later
- GitHub Copilot (for automatic capture) or direct OpenAI/Anthropic API access (for proxy)
License
MIT - see LICENSE
