Implement Databricks-Style Key Execution

[splinterhq]

Data Bricks has this really cool way of letting you begin a key with a special prefix like %lua:, %wasm: or even %python3, %m4 (you get it). So the key:

%lua:normalizeFieldVectors

Would treat whatever's in the corresponding value arena offset as runnable Lua code (it could validate first, if desired).

We can implement WASM just like Lua (I'm working on that now) and for non-embeddable runtimes, we can pass the value region contents as a file or argument and return it. To be clear this would happen on the client side only, core storage remains passive.

We can also, once I get the completion inference shard pushed, allow for %sql or other transformations using a local code model (or even third-party remote API model, or pipe through Claude code somehow).

The same sidecar methodology applies, and the results can be tied to the signal arena.

So this would make Splinter an AI "BusyBox" because it has:

• Cache
• Vector Storage
• Inference
• Executable keys
• WASM for SIMD loads
• Tandem slots for multi-modal vectors

That's cool!