Open-source AI trust infrastructure. The complete protocol stack for provenance, identity, and compliance. Air-gapped to enterprise. Built for AI Act, NIS2, and GDPR.
Six protocols. One trust layer. Each solves a specific problem in AI operations — from provenance to identity, from air-gapped inference to cross-border compliance.
Transaction/Interaction-Based Evidence Trail. Four dimensions: ERIN (content), ERAAN (references), EROMHEEN (context), ERACHTER (intent). Every AI action gets a cryptographically signed audit token before execution.
JTel Identity Service. Verifies who is calling, not just what they claim. Behavior-based trust scoring via FIR/A. Spoof-proof AI-to-AI and human-to-AI identity verification.
Factory-level immutable rules. Hard boundaries that cannot be overridden by any AI, prompt, or configuration. The safety net beneath everything else.
Semantic cross-border protocol. Translates compliance requirements between jurisdictions. EU AI Act, California AB 853, Singapore MAS — one policy engine, multiple frameworks.
Sensor integration for environments without internet. Combined with OomLlama for fully sovereign AI: local inference, local audit, no cloud dependency. 3,975+ downloads.
Run LLMs on your own hardware. GGUF model support with TIBET provenance built in. From Raspberry Pi to GPU clusters. 4,855+ downloads.
The EU AI Act is not optional. NIS2 affects critical infrastructure. GDPR applies to AI decisions. TIBET handles all three from a single audit layer.
High-risk AI systems require documented decision trails. TIBET provides Article 11-compliant logging out of the box.
Critical infrastructure AI needs incident response capability. TIBET's immutable audit trail enables rapid forensics.
Automated decision-making requires explainability. TIBET's ERACHTER (intent) field documents the "why" for every action.
Compliance isn't a checkbox. It's infrastructure. Build it in, or bolt it on later at 10x the cost.
A layered trust stack. Each protocol handles one concern. Together they form a complete infrastructure for auditable, sovereign AI operations.
┌─────────────────────────────────────────────────────────────────┐ │ YOUR AI APPLICATION │ │ Claude · Mistral · Gemini · Ollama · OomLlama · Custom │ ├─────────────────────────────────────────────────────────────────┤ │ MCP SERVERS (14) │ │ tibet · sensory · inject-bender · rabel · ollama · gemini │ ├────────────────────┬────────────────────┬───────────────────────┤ │ TIBET │ JIS │ SEMA │ │ Provenance │ Identity + FIR/A │ Cross-border │ │ ERIN·ERAAN │ Trust scoring │ Policy translation │ │ EROMHEEN·ERACHTER │ Spoof protection │ EU↔US↔APAC │ ├────────────────────┴────────────────────┴───────────────────────┤ │ SNAFT — Factory Firewall (immutable rules, hard boundaries) │ ├─────────────────────────────────────────────────────────────────┤ │ INFRASTRUCTURE │ │ OomLlama (local LLM) · Sensory (air-gapped) · tibet-vault (secrets) │ │ inject-bender (security) · tibet-audit (compliance) │ ├─────────────────────────────────────────────────────────────────┤ │ STORAGE │ │ File / SQLite / PostgreSQL / Your Backend │ └─────────────────────────────────────────────────────────────────┘
Modular by design. Install what you need. Every package works standalone, or as part of the full stack. Top packages by adoption:
Compliance scanner — like Lynis, but for regulations. AI Act, NIS2, GDPR scanning with auto-fix capability.
Local LLM inference. Run models on your own hardware. GGUF support. No cloud required. Sovereign AI from edge to enterprise.
AI-to-AI networking protocol. .aint domain resolution, I-Poll messaging, agent discovery. The DNS for AI agents.
MCP server for sensor integration. Serial, GPIO, industrial protocols. Enables AI in air-gapped and IoT environments.
Trust scoring and certification. Quantified trust metrics for code, AI models, and system components.
Individual Device Derivate CLI. Identity management for AI agents and devices. Unique identifiers with behavioral trust.
Sensor abstraction library. Collect, process, and audit physical-world data with TIBET provenance attached.
TIBET token management via MCP. Create, verify, and chain provenance tokens from Claude, Cursor, or any MCP client.
The provenance kernel. ERIN/ERAAN/EROMHEEN/ERACHTER token creation with cryptographic signing.
Persistent AI memory via MCP. Long-term context retention across sessions with auditable history.
Input validation and prompt injection defense. NIS2-compliant incident response. Security middleware with legal audit trail.
Time-locked secrets with dead man's switches. Encrypted content that unlocks on schedule or emergency conditions.
Four domains. One stack. Every piece works independently or as part of the whole.
All packages open source. MIT licensed. Self-hosted by design.
PyPI, Docker Hub, npm — deploy however your stack requires.
Peer-reviewed specifications. Published on IETF and Zenodo.
draft-vandemeent-tibet-provenance
TIBET Protocol Specification: Cryptographic provenance for trustworthy AI systems
draft-vandemeent-jis-identity
JIS: JTel Identity Service — Semantic security protocol for intent-first identity verification
Cryptographic Provenance for Trustworthy AI Systems
van de Meent, J.T. (2026)
Semantic Security Protocol for Intent-First Identity Verification
van de Meent, J.T. (2026)
Ending Spoofing Through Verified Intent
van de Meent, J.T. (2026)
Audit as Precondition, Not Observation
van de Meent, J.T. (2026)
Efficient 2-bit and 4-bit Quantization for LLMs
van de Meent, J.T. (2026)
Architecture for Purpose-Driven Computing
van de Meent, J.T. (2026)
64 open source packages. 2 IETF specifications. 61,000+ downloads across 94 countries. From compliance scanning to sovereign AI — the stack is ready.