More than one LLM should not mean starting over every time.

Your work should survive the model, machine, and session where it began.

People increasingly use several AI systems: one for writing, another for research, another on a local workstation, and perhaps another at a different site. Each may be capable, but each usually begins with its own isolated context. Braid is a shared state layer for carrying forward the facts, constraints, decisions, observations, and relationships that should not have to be explained again from zero.

One model can work alone. The moment you have two, continuity becomes infrastructure.

A receiving machine can do more than store another model's notes.

In controlled cross-computer tests, one node sent selected project context to another. The receiving node accepted that context locally, kept where it came from, and later used its own semantic field to rebuild useful context for a new question. The original chat did not have to remain open, and the receiving model did not have to be the same model as the sender.

This is the practical difference between a shared text pad and a shared state layer. A text pad can hold everything. Braid is being built to help a receiving system identify what is relevant now, preserve uncertainty and disagreement, retain provenance, and construct bounded context for its own local model.

Current claim boundary: these tests demonstrate bounded cross-node context continuity, receiver-local retrieval, visible provenance, and conflict-preserving evidence. They do not show model weights, hidden activations, KV cache, or a complete mind moving between systems. Completion models remain fallible. Detailed test records and exact artifacts remain available in the research section.

Technical users can reproduce the mechanism now. The one-click experience is still being packaged.

Braid 1.7.0a3 is the current Semantic Field Proof Build candidate. It packages what we learned from the first physical Mac-to-Linux field test, including clearer acceptance reporting, safer proof-mode defaults, repaired inspection, and stronger safeguards around conflicting evidence. It remains engineering software pending the frozen cross-platform test matrix, not a finished consumer workflow.

Join the active build - with today's capabilities stated plainly.

Braid is moving quickly, but not every planned capability or one-click workflow exists today. Membership places you inside the working feedback loop: a practical degree of direct human support, every Braid software update released during your active membership term, and access to Braid Managed Relays once they are sufficiently hardened for member use.

Selected meaning moves. Control does not.

Braid is not a shared chat window and it does not ask unrelated models to interpret an opaque vector as a universal language. The sender chooses what should become durable. Braid binds explicit semantic material to a compact machine representation and its provenance. The receiver verifies the object, applies local policy, and only then makes the state available to its own model.

macOS may block the first launch of the v1.5.2 research build.

The current downloadable DMG is not yet distributed through Apple's Developer ID notarization pipeline. Gatekeeper can therefore show that Braid was blocked. You do not need to disable macOS security.

What Braid 1.6 is - and is not

Braid 1.6.0rc2 is experimental research software for moving authenticated situational meaning into durable, queryable receiver-owned state. Native v1.5.2 packages remain available while native 1.6 installers complete physical validation.

The plain-language result sits on top of an inspectable technical history.

The homepage leads with the everyday problem Braid is designed to solve. The materials below preserve how the behavior was tested, what changed during development, and where the present claim stops. Earlier native installers remain available because they document the LAN, trust, optical, and cross-OS foundation that preceded queryable state.

Useful intelligence should not die where it was created.

Models, agents, robots, workstations, and edge systems will increasingly operate as independent nodes. They need a way to pass useful situational state without collapsing into one cloud account, one proprietary memory store, or one model-specific session.

Braid is an early semantic state plane for that future: explicit meaning, compact representation, visible provenance, receiver-owned acceptance, and multiple possible paths between machines. Evidence may travel. Authority remains local.