StegVerse is a research and engineering effort focused on building a governed distributed operating system for autonomous agents, AI systems, and human‑AI collaboration.
The project explores how complex autonomous systems can operate safely and reliably when their actions are mediated by policy enforcement, verifiable receipts, and governed state transitions.
Execution is not assumed. Execution is admitted.
Traditional software systems assume that actors can execute operations freely once authenticated.
StegVerse introduces a different model:
intent → policy gate → decision → execution → receipt → next admissible state
In this model:
- actions are evaluated before execution
- the system produces verifiable receipts
- receipts authorize subsequent actions or information access
- workflows become state‑aware and governed
Organization-crossing messages use the registered Interlock/InTr federation boundary rather than service-ID-specific transport logic. A service registered as an INTERNAL_ENDPOINT may declare an endpoint_adapter; the boundary processor dispatches that registered adapter generically after destination/service validation.
The adapter must resolve inside the organization repository root and must exist as a file. Missing, external-path, or failing adapters fail closed. Registry-driven dispatch does not itself grant transport, credential, governance, or transition authority; those authorities remain governed by their applicable layers.
The provider-neutral WorkSpace resource consumer is exposed through the registered stegverse-org.workspace-resource-consumer internal endpoint. Its organization-local adapter remains thin and delegates projection semantics to the installed canonical StegVerse SDK consumer instead of copying that logic into the organization boundary. The adapter and SDK consumer remain non-authorizing; boundary receipts, governance authority, MIR custody, and Master Records custody are not conferred by the adapter.
This keeps the organization boundary extensible without hardcoding each future service into the boundary processor.
| Component | Repo | Status | Purpose |
|---|---|---|---|
| StegVerse SDK | StegVerse-SDK | v1.0.1 | Developer toolkit for governed execution |
| Trust Kernel | Trust-Kernel | v1.0.0 | Foundational governance layer |
| StegVerse Admission | StegVerse-Admission | v1.0.0 | GCAT/BCAT admissibility evaluation |
| LLM Adapter | LLM-adapter | v2.1 | AI output governance bridge |
| Demo Suite | stegverse-demo-suite | v1.0.0 | Reproducible validation scenarios |
| Ingestion Engine | demo_ingest_engine | v1.2.1 | Orchestrated bundle ingestion |
| StegTalk | StegTalk | — | Secure messaging layer |
| StegCore | StegCore | — | Policy evaluation engine |
| Token Vault | TV / TVC | — | Ephemeral secret distribution |
The StegVerse Demo Suite provides runnable examples illustrating the core primitives:
- AI agents operate under governed execution
- actions are evaluated by policy gates
- receipts are generated and chained
- workflows unlock subsequent steps through verified state transitions
Repository: stegverse-demo-suite
StegVerse is currently in an early prototype phase, providing experimental implementations and architecture demonstrations.
Core SDK v1.0.1 is published to PyPI and integrated with the ingestion engine for automated downstream distribution.
Engineers and researchers interested in:
- AI infrastructure
- distributed systems
- autonomous agents
- governance and safety architectures
are welcome to explore the demos and participate in discussion.
Open research / prototype environment. Individual repositories define their own licenses (MIT for SDK, Trust Kernel, Admission, LLM Adapter, Demo Suite, Ingestion Engine).