Governed memory/RAG
PostgreSQL-backed memory candidates, review states, semantic retrieval, and transferability levels turn session learning into reusable organizational context instead of loose chat history.
Moebiusol Agentic Workstation Runtime
A governed AI engineering runtime combining MCPs, specialist agents, reusable skills, model teamwork, and PostgreSQL-backed memory/RAG for complex regulated work.
Open-source release
A governed AI engineering runtime for complex product, research, documentation, and infrastructure work. It combines MCP connectors, steering packs, specialist agents, reusable skills, model teamwork, and PostgreSQL-backed memory/RAG into a forkable workstation setup.
Model teamwork, not model switching
Many AI tools expose several models but still leave the user to switch between them manually. This runtime treats models as role-based collaborators: Codex/OpenAI owns execution and final synthesis, Gemini expands options and narrative structure, and Claude challenges architecture, review quality, and regulated wording.
MCP-first workstation
Connectors, skills, agents, and steering files are installed through an idempotent runtime so future forks can select the integrations they actually need.
Quality-gated delivery
CI/Qodana checks, verification scripts, public documentation, and release tags make the runtime auditable and forkable rather than a private workstation dump.
Runtime flow
Prompt
Steering
Memory recall
Model peers
MCP actions
Verification
Where it helps
The setup is designed for complex regulated product development, research consortium work, multi-repo software delivery, documentation systems, and infrastructure operations where past decisions, source evidence, and verification context must survive beyond a single prompt.