Where your compute and AI money goes. A terminal dashboard and browser GUI for AI usage, subscription limits, local inference and compute costs. Track Claude, Codex, your homelab and cloud GPUs in one ledger: tokens, GPU hours, kWh and money.
Single Python file · Standard library only · Python 3.8+ · macOS, Linux and Windows
繁體中文 · Quick start · Card previews · Manual · Release notes
Current release: v0.1.0-beta.3. Beta.3 adds a feature menu, monthly recaps and private card downloads. You can also export usage cards for social posts and GitHub READMEs.
Run computai to see what you can do: monitor, make a monthly recap,
open the browser, analyze spending, check sources or set up your environment. Choose a number and press Enter;
Enter alone opens the monitor. computai --live opens it directly.
Inside the monitor, use 1–6 or Tab to switch between overview, limits, machines, local models, spend and timeline;
h or ? shows feature help, c opens the card creator and q quits.
Run computai --web, then open http://127.0.0.1:8765/.
The browser shows the same ledger with usage charts, machine status and alerts, including a mobile layout.
Both screenshots use demo data. Use --lang zh for Traditional Chinese and --theme classic for the alternative dashboard theme.
With Tailscale installed and signed in, computai --tailscale makes the browser GUI available over HTTPS on your own tailnet;
the local server continues to listen on 127.0.0.1.
Turn your usage into an image for social posts or a GitHub README. These six previews use demo data.
| Minimal · concise banner | Terminal · console banner |
|---|---|
 |
 |
| GitHub · activity dashboard | HUD · detailed agent profile |
 |
 |
| Paper · vertical summary | Web/TUI · social sharing |
 |
 |
computai recap # open the monthly card creator; download PNG/SVG
computai --card --card-style minimal --svg card.svg # export a README cardInstall through the quick start; see cards and sharing for design comparisons, export options and automatic profile updates.
macOS / Linux — install the released beta to ~/.local/bin/computai:
curl -fsSL https://raw.githubusercontent.com/Sean-Hawks/computai/v0.1.0-beta.3/install.sh | shWindows — download and extract the ZIP from the release page, then run in that folder:
powershell -ExecutionPolicy Bypass -File install.ps1Ensure the install directory is on PATH; on Windows, reopen your terminal after installation.
From an extracted release folder, run sh install.sh on macOS/Linux or the PowerShell command above on Windows.
You can also run the single file directly with python3 ./computai (Windows: py -3 computai).
computai # feature menu; Enter opens the monitor
computai --live # open the terminal dashboard directly
computai --web # browser GUI at http://127.0.0.1:8765/Run each view in its own terminal. Claude Code and Codex usage is detected from supported local logs.
The menu lets you start monitoring without subscription questions; configure plans through
the setup option when needed. Direct --live may ask you to confirm your plans. Stop the browser server with Ctrl-C.
The menu also shows a direct command for every feature. Outside an interactive terminal,
bare computai prints the guide and exits; use computai --summary or --json for scripted usage output.
Selecting the browser option opens the local dashboard automatically; the terminal keeps the server running.
computai --setup # configure plans, machines, electricity and optional API keys
computai --doctor # inspect detected sources and missing configurationRemote machines, cloud providers and API usage are optional. Set up the sources you use.
| Source | Data | What you can see |
|---|---|---|
| Claude Code / Codex | Local session usage logs; official CLI quota queries | Tokens per model and project, cache and reasoning, subscription limits and reset countdowns |
| Gemini CLI / OpenCode / Cursor | Local recordings, usage database or exported CSV | Tokens, model shares, project breakdowns and API-equivalent cost |
| Local models | Ollama, llama.cpp, vLLM, SGLang, LM Studio and compatible servers | Loaded models, recorded tokens, inference speed and idle-model alerts |
| Machines | This computer or SSH + POSIX sh |
CPU, GPU, power, kWh and electricity cost; no remote agent required |
| Cloud GPUs | RunPod, Vast.ai and Lambda read-only APIs | Running instances, hourly costs and idle resources that are still billing |
| API organisations | Optional Anthropic and OpenAI admin APIs | Usage per model and day |
Cost analysis includes cache efficiency, month-end forecasts, budgets, subscription comparisons, GPU payback and time-of-use electricity pricing. API-equivalent values use your configured rates and are estimates, not bills. Local token coverage depends on service metrics or traffic through the optional token-counting proxy.
| Task | Command |
|---|---|
| Monthly card with PNG/SVG downloads | computai recap |
| README usage card | computai --card --card-style minimal --svg card.svg |
| Monthly usage | computai --summary --month |
| Usage by project | computai --summary --month --by project |
| Agent activity timeline | computai --timeline |
| Forecast and cost analysis | computai --analyze |
| Save an HTML report | computai --report --month 2026-09 --html september.html |
| Discover machines | computai --discover |
| Count local inference tokens | computai --proxy |
| Benchmark local models | computai --bench |
| Limit notifications at login | computai --install-watch |
| Status bars | computai --line |
Use computai --help for all options and the manual for source setup and counting rules.
The browser also exposes /metrics for Prometheus.
For usage across computers, see multi-device setup.
For tmux, SwiftBar and Claude Code, see status line integration.
Agent routing (--pick), MCP (--mcp), weekly summaries and team totals are covered in the manual.
The first run creates config.ini and prices.ini. Run computai --paths to locate your settings and SQLite ledger.
On macOS/Linux, settings default to ~/.config/computai/; Windows uses the application data directories.
Plans, API prices and electricity tariffs stay editable, with dates for checked prices.
computai --set general.lang=zh
computai --set general.theme=classicUse --setup for guided configuration or --set / --unset in scripts.
See the configuration reference for machines, secrets, tariffs and integrations.
- Imports usage fields only. Conversation contents are not retained in the ledger, reports or cards.
- Does not read
~/.codex/auth.jsonor Claude OAuth tokens. Quota queries use the official CLIs and their own login. - API keys come from environment variables or a permissions-restricted
secrets.ini(600on POSIX); they are excluded from logs, JSON, metrics and errors. - The browser server and proxy listen on
127.0.0.1by default. Cloud and admin API integrations use read-only queries. - Usage and reports remain local unless you explicitly configure sharing, remote access or publishing. Dashboard views and full exports can include project and machine names; share cards omit them.
ComputAI checks for updates at most once a day. Disable this with computai --set general.update_check=no.
Source coverage and platform limitations are documented in the beta release notes.
Beta.3 adds these features to the previous release:
| Feature | Entry point |
|---|---|
| Feature menu and monitor help | computai · h or ? inside the monitor |
| Token breakdown and local request history | Usage and tracing |
| Personal history with monthly and annual reports | Profile reports |
| Browser card creator and four new profile styles | Cards below |
For your first monthly recap, install beta.3 and run:
computai recapIt finds supported local usage, selects a month and opens the download page. Click Download PNG to share.
From the checkout, sh install.sh --recap installs and opens it in one step. First-use guide (繁中).
Create a usage card for your README, portfolio or social posts. Choose the design by its layout and the information you want to share. Cards use local aggregates and omit conversations, project paths and machine names.
| Design | Best for | Default format | Command | Available in |
|---|---|---|---|---|
| Minimal | A concise README banner with key totals | compact · 720×230 |
--card --card-style minimal |
beta.3 |
| Paper | A vertical summary with serif typography | portrait · 420×610 |
--card --card-style paper |
beta.3 |
| GitHub | Statistics, activity grid and model shares | dashboard · 900×360 |
--card --card-style github |
beta.3 |
| Terminal | A console-style banner with monospace text | compact · 720×230 |
--card --card-style terminal |
beta.3 |
| HUD | A detailed profile with agent activity, rank and badges | hud · 900×390 |
--card --card-style netrunner |
beta.1 |
| Web/TUI sharing card | Social posts, monthly recaps or a README image | Square, portrait, wide or README | recap · PNG/SVG downloads |
beta.3 |
Monthly recaps show source date coverage, record counts and comparison with the adjacent calendar month; missing records remain unknown. Supported GUI logs and recorded local-model usage can share one recap; web-only chat is outside this first version. The creator also exports aggregates for just the selected period as JSON.
With the release folder and Python 3.8+, run:
python3 ./computai recap # choose a period and download PNG/SVG
python3 ./computai --card --card-style minimal --svg card.svg
python3 ./computai --card --card-style paper --svg card.svgOn Windows, replace python3 ./computai with py -3 computai; after installation, use computai.
Add the exported SVG to your README:
The creator uses the Web/TUI design; the five profile designs use --card and export SVG.
Share the downloaded image; the creator HTML contains selectable history.
First-card guide (繁中).
In beta.3, --card-layout accepts auto (follow the style), compact, portrait, dashboard or hud.
Compact banners show key totals; dashboard and portrait add activity grids and hours.
The HUD design includes the fuller agent profile with decorative animation; the other profile designs are static by default.
computai --card --card-style github --card-layout compact --svg card.svg
computai --set card.style=paper --set card.layout=portrait # save defaults for exports and daily updatesEvery profile design supports light/dark mode (--card-theme) and English/Traditional Chinese labels.
The HUD colour presets are netrunner (default), arasaka, militech, amber, matrix and synthwave;
they share the same layout. To compare all presets and layouts locally, use computai --card --html styles.html (beta.3).
Display names, custom accents and other options are in the card reference.
computai --card --setupThe optional wizard configures your profile repository, writes the light/dark SVG pair and adds the README snippet.
Repository creation, pushing and background updates are choices in the setup.
For manual exports, use --card --svg; --card --publish PATH writes and commits the pair, and --push enables pushing.
See profile setup and the card reference.
tests/run.sh # unit tests; Python 3.8 also runs when available
python3 tests/e2e_ollama.py # optional Ollama end-to-end test; requires qwen3:0.6bReport problems through GitHub issues, including your version,
operating system, reproduction steps and computai --doctor --redact output.
MIT. Machine sampling, GPU detection, web security headers and Prometheus output derive from
slurmtop, by the same author, at commit 84cd35d.