One Workspace for Multiple AI Agents—With a Veto Button

What it does Shob runs several AI agents simultaneously inside a single project, giving each one visibility into your repository state—active branch, local changes, and all—while requiring explicit sign-off before executing commands or modifying files. It is essentially a control tower for AI assistants that refuses to let them touch the controls unsupervised.

The interesting bit The subagent delegation model turns a single large prompt into a parallel workflow: you partition the work, assign pieces to specialized agents, and watch them execute while the workspace consolidates results in real time. That shifts the interaction model from a lonely chat thread to something resembling a very small, very literal dev team.

Key highlights

• Runs multiple AI sessions in parallel against the same codebase
• Gates every file access, terminal command, and code change behind user approval
• Reads Git context—active branch, local diffs, and repository state—to keep agents aligned
• Delegates large tasks to subagents and aggregates their output live
• Surfaces terminal output and diff previews so you can inspect changes before accepting them

Caveats

• The project is explicitly in beta, so expect shifting capabilities and rough edges
• The README pitches affordability and accessibility but offers no pricing, hosting, or licensing details, leaving the actual cost and deployment model unclear

Verdict Developers who want to orchestrate several AI tools at once without surrendering their terminal should take a look. If a single chat window already meets your needs, this is likely overkill.