v1.4.1: A Desktop App and AI Session Tracking

Two things we kept hearing

Since launching the CLI and Cloud, two requests kept coming up. The first: “I love the data, but I want to see it without running commands.” The second: “I’m spending real money on AI coding agents and I have no idea how much.”

v1.4.1 addresses both. A native desktop app that gives you everything the CLI does in a visual interface. And structured AI session tracking that captures exactly which provider, model, and tools you used—along with cost and token counts—automatically, every time you commit.

Worktale Desktop

The desktop app isn’t an Electron wrapper around a web page. It’s a Tauri + React app with a Rust backend reading the same local SQLite database the CLI writes to. Native performance. 2.8 MB installer (~10 MB installed). Fast cold start.

Eight views. Full CLI parity. Everything you can do from the terminal, you can do here—plus things that only make sense with a GUI:

• • Dashboard — current & best streaks, week chart, commit timeline, 52-week heatmap with hover tooltips
• • Daily Log — day-by-day commit history with editable notes. Click any day to drill into the details
• • History — all-time stats, top modules, coding hours heatmap showing when you actually work
• • Digest Editor — full CodeMirror markdown editor. Generate digests from commit history, edit, publish to Cloud
• • AI Sessions — cost tracking, token usage charts, tool distribution, provider & model breakdowns
• • Cloud Panel — publish digests, sync with worktale.dev, manage your public profile
• • Repo Manager — add, remove, and browse all tracked repositories
• • Settings — providers, timezone, color scheme, cloud config

Ctrl+K opens a command palette. Ctrl+1 through Ctrl+8 jumps between views. It’s keyboard-first—designed for developers who don’t reach for the mouse.

---

AI session tracking

Here’s the reality of AI-assisted development in 2026: you’re spending money on it, and you have no idea how much. You switch between Claude and GPT depending on the task. You use Codex for one repo and Copilot for another. At the end of the month, you see a bill and try to reverse-engineer what happened.

Worktale now tracks all of it. What gets captured depends on which host you use:

• • Provider & model — Anthropic, OpenAI, or GitHub. Claude Opus, GPT-4o, o3, whatever you used
• • Cost in USD — automatic on Claude Code (computed from token usage against published rates), aggregated daily, weekly, monthly
• • Input & output tokens — automatic on Claude Code via transcript parsing; know exactly how much context you’re burning
• • Agent tools used — Read, Edit, Bash, Grep, Write, Agent—see which tools your AI actually reaches for
• • MCP servers — GitHub, Linear, Slack—which integrations the agent used during the session (Claude Code only)
• • Duration & commits — how long the session ran, which commits it produced
• • Narrative note — the AI writes a 1–2 sentence summary of what it built and why, after every commit

Three hosts, one CLI

Official plugins for Claude Code, OpenAI Codex, and GitHub Copilot. All three record a per-commit narrative and session metadata via worktale session add. The one difference is how much the plugin can capture on its own versus how much the agent has to supply.

You can always record sessions manually on any host (or from any agent we haven’t shipped a plugin for yet) with worktale session add.

What a real Claude Code session looks like

I activated /worktale and asked Claude Code to add rate limiting. After each commit, Claude writes a 1–2 sentence narrative. When the session ends, the SessionEnd hook parses the transcript and records the aggregate metrics automatically.

Those input/output/cost numbers aren’t guesses—the hook reads transcript.jsonl, sums every assistant turn’s usage block (including cache reads and cache writes), and multiplies against the published rate table. You get the same number you’d see on your Anthropic invoice.

The agent handles all of that. You just code. At end of day, open the desktop app or run worktale session stats and you get the full picture:

In the desktop app, this becomes interactive charts—daily cost bars, token usage over time, tool distribution pie charts, model breakdowns. Four tabs: Overview, Sessions, Models & Providers, Agent Tools.

---

Why this matters

AI spend is the new cloud bill. It creeps up. You don’t notice until the invoice arrives. And unlike AWS, there’s no Cost Explorer breaking it down by service and resource.

Worktale gives you that visibility—per session, per model, per repo. You can see that you spent $14 on Claude Opus this week but only $3 on GPT-4o. You can see that 60% of your agent’s tool calls were Read and Bash. You can see that your longest session burned 40K input tokens on a single bug fix.

This isn’t just accounting. It’s understanding how you use these tools so you can use them better.

Getting started

Desktop app

Download, install, open. If you already have the CLI, your data is already there.

The desktop app requires the Worktale CLI to be installed. The app reads the same local database—it doesn’t replace the CLI, it complements it.

AI session tracking

Plugins for Codex and Copilot work the same way. See the integration docs for install instructions for all three.

You can also track sessions manually if your agent isn’t supported yet:

Still local. Still private.

Nothing about this changes the privacy model. AI session data lives in the same local SQLite database as everything else. No network requests. No telemetry. The desktop app is a local binary reading a local file. Cloud sync remains strictly opt-in.

What’s next

macOS and Linux desktop builds are next. On the roadmap: surfacing AI session data on Cloud profiles at worktale.dev—so your public portfolio could show cost trends, model preferences, and tool patterns alongside the commit history. Today the data is captured and stored, but only visible in the CLI and desktop app.

The CLI stays free and open source. Cloud stays the paid product. Desktop is free.

---

Links:

• • Download Worktale Desktop (Windows x64 installer, 2.8 MB)
• • npm package — npm install -g worktale@latest
• • Integration docs for Claude Code, Codex, and Copilot
• • Full documentation
• • Worktale Cloud — public profiles, AI digests, standup generators