A public, MIT-licensed package for a local Codex workstation
Moebiusol Agentic Runtime
Moebiusol Agentic Runtime is a public, forkable workstation package for Codex. The public README identifies version 1.7.0. It supplies modular steering, specialist skills, an MCP connector selector and installer, optional governed memory/RAG integration and documentation checks. Download and inspect the source under the MIT licence; configure your own accounts and services for live use.
Public source available under the MIT licence. The public main-branch README identifies release 1.7.0; consult the repository for the current files and prerequisites.
Your starting point
Who this is for
- Engineering teams seeking repeatable AI-assisted development practices
- Practitioners reusing focused workflows and specialist skills
- Organisations needing to inspect sources and unresolved issues in agent contributions
Customer value
What this helps you do
Inspect and adapt the package
Public source, configuration examples, installation instructions and an MIT licence let your team evaluate the package before commissioning consulting.
Choose the integrations you need
Connector selection and optional state services let the installing team decide which accounts, knowledge sources and infrastructure to configure.
Explore the product
Product capabilities
- Modular steering, specialist roles and reusable skills for local Codex work
- MCP profiles, connector selection and repeatable installation
- Optional PostgreSQL/pgvector memory, Valkey cache and MinIO/S3 artifact integration
- Peer-model integration, documentation obligation packs and validation commands
Install steering, skills and selected connectors
The public package includes modular instructions, specialist roles and skills, an MCP registry and an installer. The installation guide provides prerequisite checks, connector profiles and generated Codex configuration snippets. Re-running installation reconciles the selected configuration; third-party accounts are not created by the installer.
Optional memory and artifact storage
For teams configuring governed retrieval, the public package documents PostgreSQL with pgvector for durable records and semantic search, Valkey for transient state and MinIO/S3 for artifact bytes. Memory integration is opt-in and requires the installing team’s services and connection settings. These are integration contracts, not included hosted services.
Inspect the installation and its evidence
Begin with the public quick start and check the selected integrations in your own environment. The repository documents test, public-scan and MCP-build commands. Optional documentation tooling can produce obligation packs and validate their source/template manifests. These tools produce review material; their presence and static badges do not prove a particular installation has passed.
- Check prerequisites: Codex access, Git, Python, uv and the documented Node.js toolchain.
- Select connectors: run make configure-mcps MCP_PROFILE=core or choose another documented profile.
- Install: run make install, then supply the accounts and optional services needed by your selection.
- Verify: follow the documented tests and inspect integration results before adopting the workflow.
Public package and paid adoption support
The public source is available under the MIT licence without a sales enquiry. Paid consulting can separately cover workflow design, connector configuration, governed retrieval, pilot evaluation and team enablement. Provider usage, account permissions and hosting remain separate arrangements. The public package’s optional extension-host contracts are described as early fixture work, not a release-grade application API.
Illustrative scenario
What a focused dependency review could save
Assumptions
- 10 engineering changes requiring initial dependency mapping.
- Assume 45 minutes per change for manual source lookup and mapping.
- Assume 15 minutes per change with structured retrieval and initial impact mapping; subsequent engineering review is additional.
10 changes × (45 − 15 minutes) / 60 = 5 hours.
5 hours difference in initial mapping effort.
Illustrative planning arithmetic, not measured performance or a promise. Validate the lookup time, mapping completeness and additional human review effort in a pilot before estimating any net difference.
Need help adopting the public package?
Start with the public source and installation guide. For paid help, describe the workflow, connector accounts, knowledge sources and review requirements you want to integrate; we can scope configuration, a pilot and team handover.
Discuss paid runtime adoption