ideas in.
PRs out.
Describe what you want. Agents build it on your machines. Only changes that pass your tests and an independent review get merged.
$ brew install partyline-sh/tap/partylinefree to start · macOS + linux
952 downloads
plan → build → ship
plan
An agent interviews you — one decision at a time — and writes tasks small enough to build. ptln plan
build
Agents pick up tasks in parallel, each in its own git worktree on your hardware. Main is never touched. ptln crank
ship
Your tests + a reviewer that didn’t write the code. Pass both → a PR. Fail either → quarantined for a human.
one binary, four tools
the assembly line
Point a fleet of agents at a backlog; review the PRs that come back.
/software-factory →session manager
Every Claude Code, Codex, and Gemini session on your machine — one searchable switchboard. Free, local, no account.
/llm-session-manager →shared context
A team memory your agents read at launch and keep current as they work.
/shared-context-for-ai-agents →multiplayer terminal
Share your shell live, end-to-end encrypted. Steer an agent together.
/share-terminal →the fine print, up front
your keys stay yours
Agents run on your machines, in your git. The server never holds a model key.
any agent
Claude Code, Codex, Gemini — plus open-weight models via the llm bridge.
encrypted end to end
Live sessions go through a blind relay. It forwards ciphertext it can’t read.
plan it together. agents build it.
ship what passes.
$ brew install partyline-sh/tap/partylinefree to start · your agents, your machines, your keys