01
Local Model (Hermes 3 via Ollama)
Zero cost, zero data leaves your computer. The model runs locally on your GPU or CPU via Ollama.
- Hermes 3 8B: Recommended default. Needs ~5.7 GB VRAM.
- Hermes 3 3B: Lightweight alternative. Fits in ~2.0 GB RAM/VRAM.
- Launch:
ollama pull hermes3:8b then select in Settings → Models.
✓ 100% Sovereign & Offline
02
Bring Your Own API Key (BYOK)
Connect directly to commercial provider APIs at raw token costs. Keys are stored on your computer in data/keys.json with 0600 permissions (not encrypted, not in OS keychain).
- Supported Providers: Anthropic, Google Gemini, OpenAI, and OpenRouter endpoints.
- Model Selection: Press List models in Settings → Models to see the models your key can use.
- Storage: Keys live strictly in
data/keys.json (0600 permissions, never sent to the portal).
▣ Monthly Cap Pre-Check Enforced
03
Sign in with X (xAI Subscription)
Use your SuperGrok or X Premium+ plan without buying separate API credits.
- Device Code Flow: The sign-in flow is built into the carrel UI.
- Status: Built but Off: Requires an official client ID issued by xAI to SemeSeam (none has been issued yet).
- Custom Client ID: Users can optionally provide their own registered xAI client ID in settings.
★ Built · Awaiting xAI Client ID
Recommended Setup
For day-to-day writing and code analysis, run Hermes 3 8B locally. For intense reasoning descents requiring frontier models, configure Claude Haiku or Gemini Flash with a $5.00 monthly cap in Settings → Spending (the carrel checks recorded month-to-date usage before each paid call and halts once reached).