tartarus-console 0.2.1

TartarusAI — Uncensored AI for builders. Developer-first console powered by your TartarusAI account, with built-in usage tracking and quota enforcement.

TartarusAI

Uncensored AI for builders.

A developer-first AI console powered by your TartarusAI account.

The CLI ships every chat through api.tartarusai.dev: the backend authenticates your token, checks your plan quota, forwards the request to the inference cluster, and writes a UsageEvent for every reply. Result — every token you spend on the CLI counts against your plan and shows up in your dashboard's Usage panel. There is no client-side path that bypasses this.

Install

One-click (recommended):

• Windows: download tartarus-setup.bat, double-click.
• macOS / Linux: curl -sSf https://dash.tartarusai.dev/tartarus-setup.sh | bash

Manual:

pip install tartarus-console

Python 3.10+. On Windows, install Git for Windows first — the CLI uses Git Bash for the terminal tool.

The package registers a single console script: tartarus.

Authenticate

You need a TartarusAI account. Sign up / log in at https://dash.tartarusai.dev.

1. Open https://dash.tartarusai.dev/account.

2. Under CLI access, click Generate token, copy the value (you only see it once).

3. In your terminal:

   tartarus auth set <PASTED_TOKEN>
   

   The CLI stores the token at ~/.tartarus/auth.json (mode 0600), verifies it against /api/auth/me, and prints your plan. If the token is invalid the local file is wiped and an error points back at the dashboard.

Sanity checks:

tartarus whoami      # email + plan + endpoint
tartarus usage       # current weekly + session token usage

Use

tartarus chat                       # interactive console, defaults to tartarus-coder
tartarus chat -z "Explain this codebase"   # one-shot mode
tartarus chat -m tartarus-thinking  # the other tracked model

That's it. No other chat invocation is supported. Specifically:

• --provider X is rejected — the CLI is locked to the TartarusAI backend (refusing this prevents inference traffic from leaking around the metering middleware).
• -m MODEL only accepts tartarus-coder or tartarus-thinking — anything else would be 400'd at the backend anyway, and we surface it early.

Every turn:

1. The console sends an OpenAI-spec request to https://api.tartarusai.dev/api/v1/chat/completions with your Sanctum CLI bearer.
2. The Laravel backend runs auth:sanctum, throttle:60,1 (rate limit per user), and the quota middleware (weekly token cap).
3. The backend forwards to the inference cluster, captures the usage block from the response, and writes a UsageEvent with source: cli.
4. Your weekly bar moves under https://dash.tartarusai.dev/settings/usage.

If the upstream omits a usage block the backend refuses to relay the response. There is no path that returns inference content without accounting.

Revoke / rotate

• From the dashboard — Account → CLI access → revoke on the token row. Effective immediately.
• From the CLI — tartarus logout revokes the token currently in use and deletes the local credentials file.

Generate a fresh token whenever you suspect a leak. Other CLI tokens on your account are untouched.

Layout

• tartarus_console/ — TartarusAI dispatcher (auth, config bootstrap, CLI commands). Locks the chat path to api.tartarusai.dev.
• The agent runtime under agent/, tools/, skills/, providers/, plugins/, and hermes_cli/ is vendored from the upstream Hermes Agent project (MIT, Nous Research) — kept unmodified except where branding or security wiring required edits. End-users only see the TartarusAI brand; the upstream module names appear only in import statements and internal docstrings.
• See README.upstream.md for the original upstream documentation.

Licence

MIT, inheriting the upstream Hermes Agent licence (see LICENSE). TartarusAI adapter code (tartarus_console/) is also MIT-licensed.