otak-usage

See what your AI pair programmers would cost - Claude Code and OpenAI Codex CLI usage as API-equivalent dollars, right in the VS Code status bar.

Usage

1. Install the extension - a cost readout appears on the right side of the status bar:

   ✦ $12.34 ⬡ $5.67 (✦ Claude Code / ⬡ Codex CLI)

2. Hover for a per-model breakdown of today and the current month

3. Click to toggle the displayed period between Today and This Month

Disclaimer: The amounts shown are API-equivalent estimates computed from your local session logs and public per-token API prices. If you use a subscription plan (Claude Pro/Max, ChatGPT Plus/Pro), this is not what you actually pay - it is what the same usage would have cost via the API.

Features

otak-usage is a lightweight VS Code extension that reads the local session logs of AI coding CLIs and turns token counts into dollars. No accounts, no API keys, no network calls.

Key Features

• Two providers, one glance:

  • Claude Code (~/.claude/projects/**/*.jsonl)
  • OpenAI Codex CLI (~/.codex/sessions/**/rollout-*.jsonl)
  • Either one alone works fine - a missing tool shows as —

• Accurate cost model:

  • Per-model pricing tables (Claude Fable 5 / Opus / Sonnet / Haiku, GPT-5.x / Codex families)
  • Claude cache writes (5m and 1h) and cache reads priced at their real multipliers
  • Claude fast mode (/fast) responses detected and billed at the fast-mode premium
  • Codex cached input priced at the cached-input rate
  • Duplicate transcript records deduplicated, session-cumulative counters handled correctly
  • Unknown models count as $0 and are flagged in the tooltip - add prices via otakUsage.pricingOverrides

• Fast and incremental:

  • Only files modified in the current month are considered; only new bytes are read
  • Scan state persists across VS Code restarts (full rescan happens once, ever)
  • Steady-state refresh takes milliseconds even with gigabytes of logs
  • Polling slows down automatically when the window is unfocused

• Remote-ready:

  • Runs in the workspace extension host, so it reads the logs of the machine where your CLIs actually run - including GitHub Codespaces, Dev Containers, and Remote-SSH
  • Windows, macOS, and Linux

Commands

Configuration

How costs are computed

• Claude Code: each assistant message records input_tokens, output_tokens, cache_read_input_tokens, and a 5m/1h cache-write breakdown. Cost = input × base + cache reads × 0.1 × base + 5m writes × 1.25 × base + 1h writes × 2 × base + output × output price. Fast-mode responses (usage.speed: "fast") are tracked as <model>-fast and billed at the fast-mode premium, with the cache multipliers stacking on the fast price.
• Codex CLI: each turn records last_token_usage. Cost = (input − cached) × base + cached × cached-input price + output × output price. Reasoning tokens are already included in output tokens.
• Prices were verified against the official Anthropic and OpenAI pricing pages. Dated model ids (claude-opus-4-8-20250915) and variants (gpt-5.3-codex-spark) resolve to their base model by longest-prefix match.

Privacy

Everything stays on your machine. The extension reads token-count metadata from local log files; it never reads your prompts' content, never writes to the log directories, and never touches the network or credential files.

Requirements

• VS Code 1.90.0 or higher
• At least one of Claude Code or OpenAI Codex CLI with existing local session logs

Related Extensions

Check out the other otak-series extensions:

• otak-monitor - System monitoring in the status bar
• otak-clock - Dual time-zone clocks
• otak-committer - AI-powered commit messages

License

MIT License - see the LICENSE file for details.

For more information, visit the GitHub repository.