Router
Chooses provider/model based on task and availability.
AI LAB / LITTLE X
The intelligence is not locked to a provider.
A persistent personal-agent research project built around memory, tools, routing, auditability and continuity.
Memory
Persistent context with selective retrieval.
Tools
Capabilities with explicit scope and audit.
Fallback
Provider failure should not mean total failure.
IMPLEMENTED
- CLI runtime + SQLite + migrations
- Local Web UI
- Ollama provider
- Fast model: qwen2.5:1.5b
- Thinking model: qwen3:1.7b
- Controlled read-only web research
- Audit and provider states
NEXT CORE
- Selective memory search
- Versioned tool registry
- Risk policy engine
- Provider routing beyond local Ollama
LATER
- Telegram
- Notion runtime
- SSH with approval
- Embeddings / PostgreSQL when justified
- Operational dashboard
Current public baseline
Phase 1D: Qwen3 Thinking, model selector and controlled web research in the local Web UI. No public authentication, external provider runtime, Telegram, SSH or autonomous sensitive execution yet.