Tally: monitor AI token usage and cost with a local dashboard and desktop companion

Tally is a free, open-source, local-first dashboard that automatically tracks and monitors token usage and cost across 47 AI coding tools — including Claude Code, OpenAI Codex, AStudio, Cursor, Gemini CLI, LM Studio, and Unsloth Studio. A desktop pet reacts to real token activity, four desktop widgets keep usage visible, and achievements turn coding history into progress. Native macOS and Windows apps sit on top of accurate cross-provider deduplication. Token counts only — never prompts or conversations.

Install in one command

Download the Tally app for macOS, Windows or Linux.

On first run, the CLI auto-detects installed AI coding tools, installs the appropriate hooks (SessionEnd hooks, TOML notify arrays, plugins, or passive log readers depending on the tool), and starts a local dashboard at http://localhost:7680. macOS users can also install the native menu bar app, available from tallyless.com.

What you get after installing

Tally runs entirely on your machine. After you install the app, a local dashboard at http://localhost:7680 exposes the following views — none of them require an account, and your data never leaves your device unless you opt into cloud sync.

• Desktop pet — a pixel companion that reacts to real token burn, streaks, and rest.
• Four desktop widgets — usage, activity heatmap, top models, and limits at a glance.
• Achievements — unlock milestones from local and optional account-wide activity.
• Unified usage dashboard — input, output, cached, and cache-creation tokens across 47 AI coding tools.
• Cost analysis — per-model pricing across 2,200+ models with an offline snapshot.
• Usage limits view — live quota and subscription status for 18 providers, with last-good caching for local providers such as Antigravity and Qoder.
• Service Status page — live operational and incident status from eight official provider status pages.
• Project attribution — see which repository or workspace consumed which tokens.
• Activity heatmap & trend charts — GitHub-style contribution calendar plus day / week / month / total / custom-range charts.
• Skills browser — discover and install agent skills from the open ecosystem.
• Native desktop apps — a macOS menu bar app and Windows system tray app.

Privacy by design: token counts only, never prompts or conversation content. Local-first; cloud sync is optional and powers only the public leaderboard.

Supported AI coding tools (47)

Use Tally as a Claude Code usage tracker, Codex usage tracker, Cursor usage tracker, Gemini usage tracker, OpenCode usage tracker, GitHub Copilot usage tracker, or DeepSeek Harness usage tracker — one dashboard covers all 47 tools below.

• Claude Code — Anthropic's official coding agent CLI; integrated via SessionEnd hook.
• Codex CLI — OpenAI Codex command-line agent; integrated via TOML notify array in ~/.codex/config.toml.
• AStudio — Codex-based coding agent; integrated via TOML notify array in ~/.acode/config.toml and passive live/archived rollout scans.
• Cursor — editor usage from local state and provider data.
• Gemini CLI — Google Gemini coding CLI; SessionEnd hook integration.
• Antigravity — passive transcript reader.
• Kiro — SQLite and JSONL readers.
• OpenCode — open-source coding agent; plugin + SQLite audit.
• OpenSquilla — passive SQLite usage reader for finalized model rows; never reads transcript content.
• OpenClaw — session plugin integration.
• Every Code — Codex-compatible TOML notify hook.
• Hermes Agent — SQLite sessions table at ~/.hermes/state.db.
• GitHub Copilot — App, CLI, and Chat usage readers.
• Kimi Code — passive wire.jsonl reader.
• CodeBuddy (Tencent) — Claude-Code fork; SessionEnd hook in ~/.codebuddy/settings.json.
• WorkBuddy (Tencent) — SessionEnd hook and passive session reader.
• WorkBuddy AI — International WorkBuddy build with separate session tracking.
• Grok Build — hook and passive cumulative usage reader.
• oh-my-pi — passive JSONL session reader.
• OmO — passive JSONL session reader, including nested subagent transcripts.
• pi — passive JSONL session reader.
• Dots — routed through pi's provider split; no separate hook.
• Prime Agent — metadata-only passive usage reader.
• Craft Agents — passive session reader.
• Reasonix — passive telemetry reader.
• Muse Code — passive session reader.
• Kilo CLI — passive SQLite reader.
• Kilo Code — passive editor task reader.
• Roo Code — passive editor task reader.
• Zed Agent — passive SQLite reader.
• Goose — passive SQLite reader.
• Droid (Factory) — passive cumulative session reader.
• DuMate — passive OpenCode SQLite reader with an independent source and reinstall-copy deduplication.
• Mimo Code — passive SQLite reader that excludes mirrored Claude history.
• ZCode — passive SQLite reader for Z.ai-native turns.
• Qoder — passive SQLite reader with exact prompt, cached-input, and completion token counters.
• AnythingLLM Desktop — passive per-message token metrics reader.
• Claude Science — passive reader for the frames table of operon-cli.db.
• DeepSeek Harness — passive session reader (~/.dsh/sessions/**/session.jsonl[.zstd]).
• TRAE Work CN — macOS signed-in app session-token reader; its internal API may change.
• LM Studio — passive local developer-server usage reader with mirrored-response deduplication.
• Unsloth Studio — passive scalar-only SQLite usage reader for local and paid-provider chat routes.
• Devin CLI — passive per-request SQLite usage reader deduplicated by request_id.
• Cline — CLI v3 / desktop app; passive ~/.cline/data/sessions transcript reader.
• MiniMax Code — passive JSONL session reader for the MiniMax desktop coding agent, deduplicated by message_id.
• Command Code — passive session reader for the cmd CLI (~/.commandcode/projects/**/*.jsonl), counting tokens and estimating cost with the shared model price table.
• TRAE — passive reader for local international TRAE IDE token usage; tracked separately from TRAE Work CN.

Who should use Tally

Tally is built for developers, founders, and engineering teams who rely on multiple AI coding assistants and need a single place to understand token consumption. Instead of checking each tool separately, you can compare usage trends by model, project, and time window in one dashboard.

Teams use Tally to answer practical questions: which model is consuming the most budget, whether a new workflow increases cost, how a Pro plan is tracking against rate limits, and how token usage shifts over days or weeks. This makes budgeting, optimization, and reporting easier without introducing heavy tracking overhead.

How it works

Each supported tool emits usage data when a session ends. Tally's parser (src/lib/rollout.js) normalizes provider-specific log formats into a unified schema: half-hour UTC buckets keyed by (source, model, hour_start). Aggregated buckets are written to queue.jsonl and served from a local HTTP API on port 7680. The React dashboard reads from those endpoints; cloud sync (optional) uploads bucketed counts to InsForge for the public leaderboard.

Frequently asked questions

How do I install Tally?

Download and open the Tally app. It detects your AI coding tools, installs hooks, and opens the dashboard.

Which AI coding tools does Tally support?

Tally supports 47 tools: Claude Code, Codex CLI, AStudio, Cursor, Gemini CLI, Antigravity, Kiro, OpenCode, OpenSquilla, OpenClaw, Every Code, Hermes Agent, GitHub Copilot, Kimi Code, CodeBuddy, WorkBuddy, WorkBuddy AI, Grok Build, oh-my-pi, OmO, pi, Dots, Prime Agent, Craft Agents, Reasonix, Muse Code, Kilo CLI, Kilo Code, Roo Code, Zed Agent, Goose, Droid, Mimo Code, DuMate, ZCode, Qoder, AnythingLLM Desktop, Claude Science, DeepSeek Harness, TRAE Work CN, LM Studio, Unsloth Studio, Devin CLI, Cline, MiniMax Code, Command Code, and TRAE.

Is Tally free and open source?

Yes. Tally runs on this device. Learn more at tallyless.com.

Does Tally collect my prompts or conversations?

No. Only token counts, model names, timestamps, and project attribution are recorded. Prompts and completions are never read or uploaded.

What does Tally track?

Input, output, cached, and cache-creation tokens per model, project, and time window. It also surfaces usage limits for 18 providers, including Claude, Codex, Cursor, Gemini, Antigravity, Qoder, Qoder CN, Command Code, Devin, and WorkBuddy.

How do I monitor AI token usage and cost across tools?

Tally is an AI token usage and cost tracker: install the app, and it monitors token consumption and spend across all 47 tools in one dashboard, with per-model pricing across 2,200+ models so you can see cost by model, project, and time window without manual tracking.

Does Tally work offline?

Yes. Tally is local-first. All parsing, aggregation, and dashboard rendering happen on your machine. Cloud sync is optional and powers only the public leaderboard.

Resources

• Homepage and install instructions
• AI coding token usage leaderboard — Claude, Codex, Cursor
• Claude IP check — exit IP reputation & risk (纯净度) score

Note: The usage dashboard, rate-limit panel, project attribution, skills browser, and widget previews are part of the local app at http://localhost:7680 after install — they are not standalone public URLs.