Your AI coding usage, in your editor.

Context budget plus Claude Code, Copilot and Codex usage, in tokens or dollars.

Install from Marketplace View source

4.2.0Latest version

32Models priced

16Chat commands

1.99.0+VS Code

See it work

Recording of the Tokalator dashboard: context budget, switching models, pinning a file, closing low-relevance tabs, a Claude Code turn arriving, and collapsing sections

Live context budget for the active model
Switch models in one tap
Pin a file to keep it relevant
Close low-relevance tabs
Claude Code turns update on their own
Collapse what you do not need

New in 4.2.0

Fixes first: turns no longer stay at 0, the meter and Turn Analytics follow the agent you use, only this workspace's sessions are counted, and a model you pick stays put instead of changing other windows.

All agents, side by side New

Claude Code, Copilot Chat and Codex in one tab: today's usage, turns, context used and model for each.

Tokalator: Compare Agents

See what your agents load New

CLAUDE.md, Claude Code memory, skills, AGENTS.md and Copilot instructions, with token counts. Click one to edit it.

Sidebar: Agent memory & instructions

Pin into your agent New

A pinned file keeps full relevance and goes into the next message of Claude Code, Copilot Chat or Codex.

Tokalator: Pin File

Session facts New

Context used, turns left, subagent share, thinking, cache share and the last 5 hours, measured from the logs. No guesses.

Sidebar: Turn Analytics

Agent instructions and checks skill New

AGENTS.md from your real commands and layout, a CLAUDE.md that imports it, and a SKILL.md that runs your checks.

Tokalator: Create Agent Instructions

A status bar you can trust New

Shows the followed agent's real context and window, and labels the open-files number as an estimate.

Status bar

Highlights

Context budget meter

Tokens used against your model's window, broken down by source.

Sidebar: Token Budget

Agent usage

Claude Code, GitHub Copilot and Codex turns, tokens and cost from their local logs, side by side in the All tab. Updates on its own.

Tokalator: Compare Agents

Tab relevance and cleanup

Scores every open tab and closes the noise in one click.

Tokalator: Optimize /optimize

Model catalog

32 current models with verified prices. Detects your Copilot model.

/model

Secure Workspace

Checks every secret file is hidden from agents, fixes gaps, then verifies.

Tokalator: Secure Workspace /secure

Code maps

GLOSSARY.md and SITE-MAP.md so agents find the right files first.

/codemap Tokalator: Generate Code Maps

Dashboard in an editor tab NewTokalator: Open Dashboard in Editor
Guided walkthrough NewTokalator: Get Started
Turn analytics/compaction /preview
Pinned files/pin /unpin Tokalator: Clear All Pinned Files
Long-context pricingTurn Cost card
Price watch and catalog syncTokalator: Price Watch Report /pricewatch
Instruction file audit/instructions
Opt-in telemetry and feedbackTokalator: Show Telemetry Data

In beta

These work and are tested, and may change as we hear how you use them. Feedback welcome.

Dependency Check Beta

Known vulnerabilities in every lockfile, any ecosystem, plus outdated packages and runtime end-of-life dates. Safe updates run only when you pick one.

Tokalator: Check Dependencies

Claude Code hooks Beta

Every hook across your settings scopes, checked against good practice. Open, remove or pause them in one place.

Tokalator: Review Claude Code Hooks

What Claude Code runs Beta

Tools, command-line programs and MCP servers from the last 7 days. Program names only, never arguments.

Sidebar: What Claude Code runs

What it tracks

Exact tokens

Claude Code · VS Code extension and CLI

Turns, tokens, cache hits and context size per turn.

Exact where recorded

GitHub Copilot · Chat in VS Code and Copilot CLI

Turns, model, prompt and output tokens from the sessions VS Code and the CLI keep.

Exact tokens

Codex · OpenAI Codex CLI

Turns, tokens, cached input and model per turn.

Show usage as API-equivalent dollars, as tokens only (for subscriptions), or both. When Copilot did not record a request's tokens, Tokalator estimates them and shows the estimated share.

MCP server and CLI

Token counts, budgets and Claude Code usage inside Claude Code and your terminal.

1
Clone and build (Node.js 18 or later):
git clone https://github.com/vfaraji89/tokalator.git cd tokalator/tokalator-mcp && npm install && npm run build
2
Register with Claude Code:
claude mcp add --transport stdio tokalator -- node "$PWD/build/index.js"
3
Or install the CLI globally and run it:
npm install -g . tokalator usage --days 7

tokalator — cli

tokalatorCLI 4.2.0

~/my-repo $ tokalator count src/core/claudeCodeUsage.pure.ts

Model:      Claude Opus 5.5
Tokenizer:  Claude BPE tokenizer (approximate)
Tokens:     2,745
Characters: 9,315

~/my-repo $ tokalator preview 182000

Now:              ~182.0K  [████░░░░░░░░░░░░░░░░░░░░]
After next turn:  ~182.8K  (18.3%)
Output reserve:   ~128.0K
Remaining:        ~689.2K  (~861 turns left)

Install

Marketplace

VS Code 1.99 or later. Paste into Quick Open (Cmd+P or Ctrl+P).

ext install vfaraji89.tokalator

VSIX

Build the package, then use Extensions: Install from VSIX.

cd tokalator-extension-vs && npm install && npm run package

From source

Clone, install and compile.

git clone https://github.com/vfaraji89/tokalator.git
cd tokalator/tokalator-extension-vs
npm install
npm run compile

Reference

Chat commands 16

Command	What it does
`/count`	Show current token count and budget status
`/optimize`	Close low-relevance tabs to free up tokens
`/pin`	Pin a file so it's always included
`/breakdown`	Show where your tokens are going
`/instructions`	List, count, and estimate tokens for instruction files in workspace
`/model`	Show or switch the active AI model
`/unpin`	Unpin a file so it returns to normal relevance scoring
`/reset`	Reset session state (chat turn counter)
`/compaction`	Show per-turn token growth and compaction recommendations
`/preview`	Preview token cost of your next message before sending
`/secure`	Scan workspace security and auto-generate AI agent settings
`/skill-gen`	Generate a checks skill (SKILL.md) from your build, lint and test commands
`/settings-gen`	Create AGENTS.md and a CLAUDE.md that imports it, from your project files
`/exit`	End the current session and save summary
`/codemap`	Generate GLOSSARY.md and SITE-MAP.md for agents
`/pricewatch`	Rate changes and their cost on your agents' last 7 days

Settings 11

Setting	Default	Description
`tokalator.relevanceThreshold`	0.3	Tabs below this relevance score are marked as distractors (0-1)
`tokalator.windowSize`	0	Override context window size in tokens. Set to 0 (default) to use the selected model's window.
`tokalator.model`	claude-opus-5-5	AI model to calculate context budget against
`tokalator.autoRefreshInterval`	2000	Dashboard refresh interval in milliseconds
`tokalator.enableSessionLogging`	false	Enable opt-in session logging for research studies. Logs only aggregate counts and scores — no file names or contents.
`tokalator.claudeCode.enabled`	true	Show token usage and cost from local Claude Code session logs (CLI and VS Code extension) for this workspace. Reads ~/.claude/projects (or CLAUDE_CONFIG_DIR) on this machine; nothing is uploaded.
`tokalator.agents.copilot.enabled`	true	Show GitHub Copilot usage: Copilot Chat sessions that VS Code stores for this workspace, and Copilot CLI sessions started in this workspace. Read locally; nothing is uploaded.
`tokalator.agents.codex.enabled`	true	Show OpenAI Codex CLI usage for sessions started in this workspace, read from local Codex session logs (CODEX_HOME or ~/.codex). Nothing is uploaded.
`tokalator.costDisplay`	dollars	How Claude Code usage and turn costs are shown. Subscriptions are not billed per token, so subscribers may prefer tokens.
`tokalator.telemetry.enabled`	false	Share anonymous daily usage counts (feature use, model IDs, Claude Code totals) with tokalator.ai to improve the extension. Off by default. Never includes code, file names, paths, or prompts. Also requires VS Code telemetry to be on. Use Tokalator: Show Telemetry Data to see exactly what is sent and Tokalator: Delete My Telemetry Data to erase it. Privacy policy
`tokalator.dependencyCheck.auto`	false	Check every lockfile for known vulnerabilities and outdated packages once a day, and warn about critical or high issues or an end-of-life runtime. Package names and versions are sent to OSV.dev, the package registries and endoflife.date. Off by default.

Tokenizers 3

Provider	Models	Tokenizer	Notes
Anthropic	Claude Fable, Mythos, Opus, Sonnet, Haiku	`@anthropic-ai/tokenizer`	Approximate for Claude 3 and later; Claude Opus 4.7 and newer tokenize the same text into roughly 1.0 to 1.35 times as many tokens
OpenAI	GPT-6, GPT-5.6, GPT-5.5, GPT-5.4, GPT-5.3-Codex	`o200k_base (js-tiktoken)`	OpenAI's published encoding
Google	Gemini 3.8, 3.6, 3.5 Flash, 3.1 Pro, 3 Flash, 2.5 Pro	`~4 characters per token`	Heuristic; Gemini's tokenizer is not published for local use

Questions

Does Tokalator track Claude Code usage?

Yes. It reads Claude Code's local session logs, written by both the VS Code extension and the CLI, and shows turns, requests, cache hit rate, context size per turn, and cost. It refreshes every 15 seconds while the sidebar is open, and the active model follows the agent you used last.

Does Tokalator keep any of my data?

Zero data retention for your work. Your code, prompts, chats, file names and paths never leave your machine, so nothing about your work is stored anywhere. With telemetry off, which is the default, the extension sends nothing about you to any server. If you opt in, only anonymous daily counts are kept, deleted after 13 months or whenever you ask.

Does it track GitHub Copilot usage?

Yes, from version 4.1.2. VS Code keeps your Copilot Chat sessions on your machine, and recent versions record each request's prompt and output tokens and the model that answered. Tokalator reads those files locally, estimates the requests where tokens were not recorded, and shows the estimated share. It also reads Copilot CLI sessions. For multi-step agent requests VS Code records only the final prompt size, so input can be undercounted.

Which coding agents does it support?

Claude Code (VS Code extension and CLI), GitHub Copilot (Chat in VS Code and the Copilot CLI) and the OpenAI Codex CLI. Each has its own tab on the usage card and its own setting. Everything is read on your machine.

What does API-equivalent cost mean?

It is what the tokens would cost at API list prices. If you use a Claude, Copilot or ChatGPT subscription you are not billed per token, so switch the usage card to Tokens (setting tokalator.costDisplay) to hide dollar amounts.

Is my code sent anywhere?

No. Everything runs locally. The only default network request downloads the public pricing catalog from GitHub. Telemetry is opt-in, off by default, and never includes code, file names, paths, or prompts.

Is Tokalator free?

Yes. The extension, the website tools, and the MCP server are free and open source under the AGPL-3.0-or-later license.