Single-Core Philosophy
Why LISA's identity is independent of any model or channel.
The single-core, multi-shell architecture exists because LISA's identity and capabilities are independent of any language model or interaction channel. The core resides on the server and is accessible via REST from any shell. Shells differ only in four dimensions:
- Model — different inference engines executing the same personality and protocols. This provides redundancy: if one model degrades, the other continues with identical OS capabilities.
- Availability — the server shell runs continuously; the local shell runs on demand.
- Hardware-bound tools — some tools require local hardware (for example a QR-authenticated messaging bridge used for context ingestion, or a design-tool token). These are deployment constraints in the tool manifest, not architectural boundaries.
- Interaction channel — the channel determines UI affordances, not OS capabilities.
Interchangeability made concrete
The shared memory gateway makes interchangeability real. Both shells read from and write to the same datastore over REST, so a dispatch initiated on one channel is visible and continuable on the other, and cache entries written on one are assembled by the other. Sessions are inter-aware through a cross-channel session registry.
The consequence: there is no "primary" shell. There is one LISA, reachable through whichever door is convenient.