Explore more publications!

Realflow.ai Releases Prompt State Protocol (PSP) for LLMs as Open Standard Under Apache 2.0 License

PSP The embedded tag language for the context window.

Realflow.ai Logo

First open specification that enables AI workflows directly in the context window while addressing critical prompt injection vulnerabilities.

We're not claiming to have solved prompt injection, but PSP moves the defense to a cryptographically-secure layer.”
— Mick Seals, CTO
HOUSTON, TX, NY, UNITED STATES, December 9, 2025 /EINPresswire.com/ -- Realflow.ai Releases Prompt State Protocol (PSP) as Open Standard Under Apache 2.0 License
First open specification that enables AI workflows directly in the context window while addressing critical prompt injection vulnerabilities

Energy LIVE 2025 Conference - Realflow.ai today announced the public release of the Prompt State Protocol (PSP) specification under the Apache 2.0 license, making enterprise-grade AI workflow orchestration and prompt security freely available to developers, researchers, and organizations worldwide.

PSP represents a new approach to building AI-powered applications. Rather than requiring external orchestration engines or custom programming, PSP enables complete business workflows to execute directly within an AI model's context window-guided by cryptographically signed instructions that the model reads and follows like natural language.

"Think about how a good business conversation works," said Mick Seals, Founder and CEO of Realflow.ai. "A single thread runs through it. You don't jump randomly between topics-there's a natural flow from discovery to decision to action. PSP brings that same structure to AI interactions. It helps guide business conversations along defined paths while maintaining focus, even across multiple sessions or human approval steps."
________________________________________
Addressing the Prompt Injection Challenge
Prompt injection-where malicious input manipulates AI system instructions-remains the top security vulnerability in AI applications according to OWASP. PSP addresses this through cryptographic signatures that create verifiable trust boundaries. System instructions are signed and timestamped, allowing AI models to distinguish between trusted governance rules and untrusted user input.

"We're not claiming to have solved prompt injection," Seals noted. "But PSP moves the defense from the semantic layer-where attacks consistently succeed by rephrasing-to the mathematical layer, where cryptographic verification provides deterministic protection."
________________________________________
Zero Trust
We have found a way for the LLM to recognize when an attacker is trying to use agents in an inappropriate way. We default to zero trust and then allow individual steps in the conversation (we call them nodes) to be associated with only the agents it may need.
________________________________________
Workflows Without the Wiring
Beyond security, PSP introduces a fundamentally different way to build AI applications. Traditional approaches require developers to write orchestration code that drives the AI step by step. With PSP, the AI model itself becomes the orchestration engine, reading workflow definitions as natural language and executing them autonomously.

This enables business analysts and domain experts to define sophisticated multi-step processes-complete with decision points, human approvals, and external integrations-without writing code. The same PSP workflow runs unchanged across Claude, GPT, Gemini, or any capable model.
________________________________________
Open Standard, Commercial Services
The complete PSP specification is available immediately at realflow.ai under Apache 2.0 licensing. Organizations can implement PSP in their own infrastructure at no cost.

Realflow.ai will follow this release with a complete SaaS platform scheduled for January 1, 2026, offering:
• Visual workflow builders for creating PSP applications
• Hosted signing API services with key management
• MCP servers for signature verification and workflow persistence
• Pre-built connectors for enterprise systems
• Human-in-the-loop approval workflows with expiring links and notifications
________________________________________
Get Started Today
• Download the Specification: Full RFC documentation available at https://github.com/realflowmick/psp
• Try It Now: Download the sample system prompt, an example PSP workflow, sign it, and run it.
• Register for Whitepapers: Technical deep-dives on PSP architecture and implementation patterns
• Join the Early Access Program: SaaS platform access granted first-come, first-served beginning January 1, 2026
To register your interest and receive whitepapers, visit realflow.ai/early-access.
We are excited to see what you build!
________________________________________
About Realflow.ai
Realflow.ai is building governance-first AI infrastructure for the enterprise. Founded by Mick Seals, a 40-year veteran of enterprise consulting, Realflow.ai develops open standards and commercial services that enable organizations to deploy AI applications with confidence. The company is headquartered in Houston, Texas.
________________________________________
Media Contact
Mick Seals
Founder & CEO
https://realflow.ai
press@realflow.ai
________________________________________
PSP is an open specification released under the Apache 2.0 license. Apache 2.0 is a permissive license that allows commercial use, modification, and distribution.

Mick Seals
Realflow.ai
+1 315-359-9198
mick@realflow.ai
Visit us on social media:
X

Legal Disclaimer:

EIN Presswire provides this news content "as is" without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the author above.

Share us

on your social networks:
AGPs

Get the latest news on this topic.

SIGN UP FOR FREE TODAY

No Thanks

By signing to this email alert, you
agree to our Terms & Conditions