Current status: Pre-v0.1 — Architecture and documentation phase.
The full milestone details, module specifications, and design decisions live in the PRD and Architecture Doc. This file is the quick-reference version.
Goal: Prove the core loop works. Send a message via curl, get an AI response back.
- Pi harness container running, connected to LM Studio on Mac Mini
- Sentinel Core container routing messages to Pi
- Basic
docker-compose.ymlstructure in place
Goal: Obsidian vault integration. The system reads context before responding and writes session notes after.
- Obsidian Local REST API plugin connected
- Core retrieves user context from vault before building Pi prompt
- Session summaries written to vault after each conversation
- Existing Obsidian data imported to
/inbox/imports/
Goal: First real interface. Talk to the Sentinel without a terminal.
- Discord bot container OR Apple Messages bridge operational
- Standard message envelope format finalized
- Docker Compose override pattern validated
Goal: Robust, swappable AI provider configuration.
- Provider config via environment variables only
- At least two providers testable (LM Studio + one other)
- Error handling, retries, timeouts
- Pi harness API contract finalized — everything else builds on this
First real module. Proves the pluggable module architecture works.
- NPC roster management (create, update, query)
- Session note capture and world state tracking
- Dialogue generation on demand
- Delivered as a Docker Compose override file
Simpler second module. Validates that the module pattern is repeatable.
- Log practice sessions via Discord or Messages
- Query practice history with natural language
- Chord/melody idea capture
AI-assisted development environment for building new Sentinel modules.
- Separate Pi harness instance tuned for code tasks
- Routes heavy tasks to a more capable cloud model (e.g., Claude API)
- Scaffolding generator for new modules
- Isolated from production Sentinel
OFX transaction import and spending intelligence.
- OFX file parsing and deduplication
- AI-assisted transaction categorization
- Budget tracking and natural language spending queries
- Recurring charge detection
- Monthly summary reports in Obsidian
AI trading agent in simulated mode, full audit trail.
- Alpaca paper trading API connected
- Personal trading rules file (plain English, you write it)
- Watchlist research loop with thesis notes
- Hard limits enforced: no margin, position size caps, daily trade cap, PDT tracking
- Complete trade rationale written to Obsidian before every execution
Real money, only if paper trading results warrant it.
- Live Alpaca API keys configurable separately from paper keys
- Optional human approval step before each trade
- Emergency stop command
- Weekly P&L summary delivered via interface
- Full documentation pass for external contributors
- Module development guide (CONTRIBUTING.md) polished
- GitHub repo structured for open contribution
- Discogs / ListenBrainz integration (if music module has proven useful)
- Foundry VTT integration investigation begins
These aren't on the roadmap yet but are worth keeping in mind as the architecture evolves:
- Foundry VTT integration — receive real-time combat events, push NPC reactions
- Media Discovery — ListenBrainz listening history → Discogs wantlist automation
- Telegram / WhatsApp / Slack interfaces — each as its own drop-in container
- Voice interface — interesting long-term direction, not v1 scope
- Multi-device vault sync — iCloud vault sync before considering Obsidian Sync
For full specs on any milestone, see the PRD.