Intersignal Braid v1.5.2 Enables Local-First AI State Transfer Across OS and Models

Independent AI research lab Intersignal has demonstrated Braid v1.5.2, a local-first protocol for authenticated semantic state transfer between AI systems. The system enables local AI instances to exchange signed data across different operating systems and embedding models while maintaining strict operator control.

The protocol utilizes "Semantic Capsules," saved as .brad objects, which bundle content with source model digests, payload binding, and provenance. Intersignal validated cross-OS capabilities, confirming successful state transfer paths from macOS to Linux, macOS to Windows, and Windows to Linux. To ensure reliability, the lab tested multiple transport methods, including direct LAN and optical QR reels; the latter was verified at 30/30 chunks during Mac-to-Mac testing. The system integrates with local Ollama loopback paths and explicitly rejects cloud markers.

The Shift to Local Interoperability

Sharing context between AI systems typically requires manual text copying or routing sensitive data through centralized cloud platforms. This creates a dependency on third-party providers and introduces privacy risks. Intersignal is developing Braid as a coordination layer to enable "local-first" AI interoperability. In this framework, trust is managed through public keys and frozen semantic routes rather than centralized accounts, allowing systems to communicate without a middleman.

Solving the Model Gap

One of the primary technical hurdles in AI interoperability is the difference between embedding models; a vector from one model is typically gibberish to another. Braid v1.5.2 addresses this by utilizing receiver-local native re-embedding. When the receiving system detects that model digests differ, it re-processes the content locally, avoiding the inaccuracies of "vector-to-text guessing." This ensures that the receiving machine retains final authority over state acceptance and interpretation.

Implications for Decentralized AI

This development signals a move toward decentralized AI infrastructure. By enabling secure, authenticated state transfer between heterogeneous local models, Braid reduces the industry's reliance on cloud providers and enhances privacy. According to Intersignal's thesis, AI systems should be able to exchange useful state without surrendering identity, provenance, or operator control.

What's Next

As Braid evolves, the focus remains on expanding the versatility of .brad objects and refining the speed of local re-embedding. While the current demonstration proves the viability of cross-OS and cross-model handoffs, the broader adoption of such a coordination layer will depend on how easily it integrates with a wider array of local LLM runners beyond the current Ollama integration.