Some of it runs today. Some of it is being built. Each answer says which.
Every organization can own the learning loop that encodes its institutional knowledge.
Live in v3: a tenant-scoped memory graph with branches, commits, bookmarks, and recall. It preserves decisions and working context outside any model's weights.
Switch the underlying model without losing the domain expertise embedded in your systems.
MCP is the transport boundary. The same Boswell tenant is used from Claude Code and Codex today; each new client gets a supported, tested connection as it is added.
Companies must retain sovereignty over their proprietary learning systems and institutional knowledge.
Live in v3: tenant-scoped credentials and row-level database controls keep each tenant's graph its own.
Private evals should capture whether a model is actually improving against outcomes that matter to the business.
Today and next: v3 keeps a typed, auditable memory history. Boswell 6 adds a provenance gate — grounded, unsupported, contradicted, or quarantined — that is being measured before it ships, and the benchmark card comes with it.
A world where value concentrates in a few models while industries are commoditized and hollowed out is unstable.
Boswell keeps memory outside model weights. That separation is what makes portability and organizational control possible, and ownership, export, and deletion are written down as product policy rather than implied.
Platforms should enable more value on top than is captured inside.
The intended role is infrastructure. Boswell supplies memory and continuity beneath agents and applications. That platform shape is the direction; today it is an invitation-only beta.
Human capital and token capital must compound together; you can offload tasks, never the learning itself.
Boswell is designed to keep the why. Human decisions, reasons, corrections, and model work can share a versioned substrate instead of being stranded in separate chats.
