J.A.N.E

JANE is a platform for orchestrating AI agents that perform real software development tasks inside actual repositories.

Instead of just generating code, JANE executes work end-to-end. Tasks move through a structured workflow — from planning to development to pull request — with agents generating specs, writing code, running commands, and producing real outputs.

During the planning phase, agents create structured artifacts like spec.md, design.md, and tasks.md. Then, development agents run on connected local workers, cloning repositories, creating isolated branches, installing dependencies, and implementing changes directly in the codebase.

JANE supports multiple AI runtimes (such as Codex, Claude, and Gemini), enabling flexibility and avoiding vendor lock-in. Every execution is fully traceable, with diffs, logs, tests, screenshots, and metadata.

By combining structured planning, real environment execution, and PR-based workflows, JANE transforms product requirements into shipped code — acting as an autonomous development layer for modern teams.