Sovereign AI Agent System
A local-first AI command center. Talk to Timmy, coordinate your swarm, gate API access with Bitcoin Lightning — no cloud AI, no telemetry, no compromise.
Capabilities
Built for the operator who wants sovereignty without sacrificing capability. Every subsystem works offline.
Agno-powered conversational agent backed by Ollama (llama3.2 default) or AirLLM for 70B–405B models on Apple Silicon. SQLite memory, no cloud.
Agno · Ollama · AirLLMFastAPI backend with Jinja2 + HTMX frontend. Dark terminal aesthetic. Real-time health polling, chat history, live swarm feed — no JavaScript framework.
FastAPI · HTMX · Bootstrap 5Spawn sub-agents, post tasks, run 15-second Lightning auctions. Registry, manager, bidder, and coordinator — all in-process or as separate subprocesses.
Coordinator · Registry · AuctionBitcoin Lightning payment gating via HMAC macaroons. Mock backend for dev, LND backend scaffolded for production. Auction bids priced in sats.
L402 · Macaroon · BOLT11Pattern-matched NLU detects intents (status, swarm, task, help) from natural language. TTS via pyttsx3 speaks responses aloud. Fully offline.
NLU · pyttsx3 · Intent DetectioniOS safe-area, 44px touch targets, 16px inputs (no zoom), momentum scroll, dynamic viewport height. 21-scenario HITL test harness built in.
iOS · HITL · ResponsiveReal-time swarm events — agent joins, task posts, auction bids, completions — broadcast over WebSocket to the live dashboard.
WebSocket · Live EventsXSS prevention via textContent (not innerHTML). Startup warnings for default secrets. HMAC-signed macaroons. Graceful degradation on every optional service.
XSS-safe · HMAC · WarningsContinuous test runner polls pytest every 60 seconds and alerts on regressions. Runs alongside dev work in the background — silent when green.
pytest · self-tdd · CISystem Design
A layered local stack. Every dependency can degrade gracefully — Ollama offline? Error message. Redis missing? In-memory pub/sub. pyttsx3 absent? TTS is a no-op.
Browser / Phone │ HTTP + HTMX + WebSocket ▼ ┌─────────────────────────────────────────────────────────┐ │ FastAPI (dashboard.app) │ │ routes: agents · health · swarm · marketplace · voice │ └────────┬──────────────┬────────────────┬────────────────┘ │ │ │ ▼ ▼ ▼ Jinja2 Timmy Agent Swarm Coordinator Templates │ ├─ Registry (SQLite) (HTMX) ├─ Ollama ├─ AuctionManager (15s bids) └─ AirLLM ├─ SwarmComms (Redis / mem) (70B–405B) └─ SwarmManager (subprocess) │ ├── Voice NLU + TTS (pyttsx3, local) ├── WebSocket live feed (ws_manager) ├── L402 Lightning proxy (macaroon + BOLT11 invoice) ├── Push notifications (local + macOS native) └── Siri Shortcuts API (iOS automation endpoints) Persistence: timmy.db (Agno memory) · data/swarm.db (registry + tasks) External: Ollama :11434 · optional Redis · optional LND gRPC
Get Running
Five minutes from zero to a running agent. You need Python 3.11+ and Ollama.
git clone https://github.com/AlexanderWhitestone/Timmy-time-dashboard.git cd Timmy-time-dashboard
make install # or manually: python3 -m venv .venv && source .venv/bin/activate pip install -e ".[dev]"
# In a separate terminal: ollama serve ollama pull llama3.2 # ~2 GB, one-time download
make dev # Opens at http://localhost:8000 # Mobile: http://<your-lan-ip>:8000
make test # full test suite — no Ollama needed make test-cov # + coverage report make watch # self-TDD watchdog in background
Contributors
This repo is built by a multi-agent team. Each tool has a defined lane. See AGENTS.md for the full development standards.
What's Next
Three phases, each named after a book.
Agno + Ollama + SQLite + FastAPI dashboard + HTMX + full test suite. CLI, mobile layout, Bootstrap, CI/CD, AirLLM big-brain backend.
Swarm coordination, L402 Lightning payment gating, voice NLU/TTS, marketplace, WebSocket live feed. Agent personas: Echo, Mace, Helm, Seer, Forge, Quill. Real LND gRPC backend.
Bitcoin Lightning treasury — agents earn and spend sats autonomously.
Single .app bundle for macOS (no Python install).
Federation between Timmy instances.