Media
1 item
Description
Track and visualize your AI model usage with a GitHub-style heatmap. Track and visualize your AI model usage with a GitHub-style heatmap. v0.2.5 — token-accounting accuracy fix • Fix: corrects several underlying token-counting bugs that significantly under-reported usage. v0.2.4 — Codex support • New: OpenAI Codex data source. v0.2.3 — Streaming protocol updates for 5 platforms • Updated stream parsers to track upstream protocol changes, restoring capture accuracy v0.2.2 — Gemini CLI support + Day Detail polish • New: Gemini CLI data source. Optionally read ~/.gemini/tmp/<project>/chats/*.jsonl session logs to capture terminal Gemini usage. Off by default — activate from the "Local Log Sources" card on the dashboard. • Improvement: DayDetail "Model" column widened so long ids like `deepseek-v4-flash-thinking` display in full instead of being truncated. v0.2.1 — Claude Code log source • New: Claude Code data source. The dashboard can now optionally read ~/.claude/projects/ JSONL session logs via the File System Access API, so terminal / IDE usage that bypasses the browser is finally captured. Off by default — only activates after you explicitly authorize a directory in the dashboard's "Claude Code Data Source" panel. Sync runs incrementally every 2 minutes; one-tap Disconnect anytime. • New: Browser / Claude Code / All filter pills above the heatmap. • Fix: DeepSeek model labels updated to v4-pro / v4-pro-thinking / v4-flash / v4-flash-thinking to match the official V4 release (detected via response stream metadata).. Privacy & permissions: • New permissions: "alarms" (schedule the 2-minute log sync), "notifications" (open the dashboard from a one-tap recovery prompt when File System Access permission lapses, e.g. after a Chrome restart). No marketing or content notifications. • Data stays 100% local. Log files are parsed in-process; raw prompts and replies are never persisted — only counts and metadata. Nothing is transmitted off-device.