Three launches make agent coding sessions shared, parallel, searchable

Mpai, Murmell and Inventory each turn a solo AI coding session into something different: shared, parallel, or searchable.

An agent session normally lives on one machine, in one window, and disappears once the terminal closes. That limits who can watch it work, how many agents can touch a codebase at once, and where months of prompt history end up.

Mpai is MIT-licensed and sits at #12 with 93 upvotes. It lets a teammate join an in-progress Claude Code or Codex session over Tailscale, seeing every turn and typing name-attributed prompts of their own. The security model stays narrow on purpose. The host Mac keeps execution authority: no arbitrary shell access, no way to bypass an approval prompt.

Murmell, #13 with 90 upvotes, takes a different angle: run more agents, not more people. It puts Claude Code, Codex, Kimi and OpenCode on the same repo inside one browser canvas. File claiming keeps them from overwriting each other's edits.

Inventory, #17 with 85 upvotes, skips collaboration entirely and solves storage. It builds a local-only search index across years of history from Cursor, Claude Code, Zed, Codex and Kiro. No signup, one-time fee. The argument: agent transcripts hold client context nobody should have to upload just to search it.