What's happening at IOBOXX
Updates from the IOBOXX programme.
- Architecture
Memory and model share the same primitive.
EAV → KV → tensor is one decomposition chain. The node attribute map is a sparse tensor row — the same primitive every AI model speaks natively — so the memory layer and the inference layer no longer need two adapters bolted together.
Read the brief → - Portability
Switch the model. Keep the memory.
IOBOXX speaks MCP — the open Model Context Protocol — so the memory layer is not coupled to any single LLM. Any MCP-aware agent can read and write the same typed memory; rotate the runtime and the rest of the platform stays intact.
See how → - Senses
Wired into memory, not imported beside it.
Upstream systems become traversable nodes in the same knowledge graph. Adopt a sense once, derive its typed taxonomy from the upstream schema, and every subsequent agent call queries through that locked shape — no staging tables, no copy drift.
Architecture / senses → - Twins
Five sovereign brains running today.
FORUS Digital, GOODFORUS, CSP, STI, and ECA — each on its own IOBOXX node, with its own grants, its own commitlog, and its own data residency. Same substrate underneath, separate sovereignty on top.
See the twins → - Open source
AGPL-3.0 core, sovereign self-host from day one.
Clone the repo, publish the SpacetimeDB module, point Claude Code or Cursor at the MCP server, and a Fundi can start writing real working memory in under ten minutes. No closed-source companion service quietly carrying the load.
Quickstart → - Contributor
Every reducer call is provenance.
The hash-chained commitlog is the proof substrate for contribution attribution. Reducer calls credit Fundi Points to the calling principal today; at TGE those Points convert to Fundi Tokens on Base for on-chain settlement.
Tokenomics →
More updates
The full programme feed will land here as the publishing cadence picks up.