Five MIT-licensed products make AI-assisted engineering inspectable across instructions, reusable workflows, isolated execution, document review, and pull-request delivery.
Execution isolation, document review, instruction distribution, engineering workflow and closed-loop delivery each have a separate product surface.
Runs Claude Code, Codex, Gemini and other coding agents inside isolated Docker Sandbox microVMs, with a repository-specific configuration committed beside the code.
Adds GitHub-style inline review comments to Markdown in VS Code. Feedback stays local, travels with Git and exports as AI-readable context for the next editing pass.
Builds, versions and distributes AI coding rules, agents and skills from one project source. The CLI manages the manifest, lockfile, package store and native output for each enabled AI tool.
Publishes the engineering rules, agents and skills used by the practice as installable Intelligence Packages. Core covers engineering and Git discipline; Spec adds the spec-driven delivery lifecycle.
Runs the closed loop across registered repositories: discovers issues, plans and implements changes, verifies the result, and delivers each change through a branch and pull request.
Each project started with a constraint in the practice or in a client repository. Sandbox Console isolates coding agents from the developer machine. Markdown Review Comments keeps document feedback local, Git-native and readable by AI tools. Intelligence keeps AI instructions versioned and reproducible across tools. Intelligence Dev Packs carries the reusable engineering layer. Operator closes the loop from issue discovery through a verified pull request.
Together they form the agentic harness: everything around the model that controls what it may do. Context and instructions, tools and access rights, the task execution loop, and checks on the result.
Client-specific rules and skills stay in the client repository, while the shared packages remain inspectable before an engagement starts. Each product page links to its public source. Installation and release details stay with the project repository or marketplace.
A Tier 0 assessment maps the current delivery system. A Tier 1 Foundation Pack installs the approved intelligence layer in one repository.
