Intersignal Demonstrates Cross-Model Semantic State Transfer Between Local AI Systems

Braid v1.5.2 moves authenticated context between heterogeneous local models while preserving provenance, receiver authority, and native representation

FORT LAUDERDALE, Fla., Aug. 17, 2026 /PRNewswire/ -- Intersignal, an independent artificial intelligence research lab and systems engineering initiative, today announced the successful physical testing of Braid v1.5.2, its local-first protocol for authenticated semantic state transfer between AI systems.

The milestone proves that sovereign AI agents can share complex context across a local network without relying on centralized cloud APIs or homogenous model architectures. In recent two-node testing on separate Apple Silicon Macs, Braid successfully transferred signed semantic context from one local Ollama embedding model to another. The source node utilized all-minilm:latest, while the receiving node utilized nomic-embed-text:latest.

Rather than attempting to translate one model's opaque vector coordinates directly into another's representation space, Braid carries explicit signed semantic material, source-model provenance, and a cryptographically bound source representation. The receiving system verifies that evidence and, when the models differ, re-embeds the authenticated material locally using its own model.

In the heterogeneous-model test, the receiving Mac produced a native 768-dimensional Nomic embedding from material originally represented in a 384-dimensional source space. The transaction completed through Braid's Phase B finality boundary with bridge_semantic_invention: false, preserving a strict distinction between sender-signed evidence and receiver-derived state.

A separate exact-space test demonstrated the complementary path. When both nodes used the same embedding model with the identical model digest, Braid triggered its exact_space_fast_path, retained the authenticated 384-dimensional representation, and bypassed unnecessary receiver-side re-embedding.

Together, the tests validate Braid's two-path semantic rule:

• Same model space: Preserve the authenticated representation directly.
• Different model space: Preserve the authenticated meaning and allow the receiving model to reconstruct its own native representation locally.

"This is not hidden-state transplantation and it is not vector-space guesswork," said David Seaman, Founder and CEO of Intersignal. "Braid is designed to move authenticated continuity between systems that may organize meaning differently. The sender proves what it meant; the receiver remains sovereign over how that meaning is represented locally."

To demonstrate the protocol's resilience in extreme, air-gapped environments, Intersignal also completed a physical optical transport test. A signed Braid object was divided into 30 QR frames, displayed on one Mac, captured by the camera of another, and successfully reassembled byte-for-byte. This optical transport layer carries the exact same signed .brad object used over conventional network transport, without requiring reinterpretation or re-signing.

Braid v1.5.2 supports Ed25519 signatures, source-model digest binding, freshness and replay checks, local receiver finality, heterogeneous Semantic Capsules, exact-space fast-path handling, LAN transport, and human-visible QR transport. Local Ollama integrations remain strictly receiver-controlled, with embedding and optional completion models operating on the user's own hardware.

The project's architecture provides a foundation for research into local-first AI memory, sovereign agent systems, offline coordination, and portable checkpoints across heterogeneous inference environments.

Braid v1.5.2 for macOS is available from Intersignal.org.

About Intersignal Intersignal is an independent artificial intelligence research lab and systems engineering initiative focused on sovereign AI infrastructure, local-first computation, model interoperability, and resilient machine intelligence systems.

Media Contact:
Missy Feldman, Intersignal
[email protected]
https://intersignal.org 
720-688-6110

SOURCE Intersignal