ThinkRail gives each coding agent its own git worktree: parallel edits stay out of your checkout, and finished workspaces shelve themselves
Synopsis
ThinkRail wraps a coding agent in a full IDE, and each task gets its own workspace: a git worktree cut from the repository, with its own branch and working directory, so several agent sessions can run at once without touching the developer's checkout. Agents can also be pointed at the project folder itself, or existing worktrees attached. Dormant workspaces settle onto a per-project shelf when their pull request merges or closes, or after a quiet week, and return when you work in them again; sessions carry their own model and a live token and cost readout.
Interpretation
The unit of code change moves from one checkout to one workspace: any git repository opens as a project, workspaces are cut from it, and an agent only touches its own. Working without a workspace stays available, since the agent can edit the project folder itself, or an existing worktree can be attached. The split is a git worktree, a second working directory on the same repository with its own branch, so several pi sessions run side by side, each on its own branch; the good branch is merged and the rest are deleted.
Agent output lands in a workbench you can review: editor tabs, a file tree, terminals and documents are scoped to the active workspace, and comments left on files and diffs travel into the chat as structured context. Intent is kept across sessions by a spec graph, a connected set of SPEC.md files beside the code that the agent reads, searches and maintains through dedicated spec_* tools. A workspace can run several pi sessions at once, each with its own model and a live token and cost readout, and each completed step of a shared plan can be committed and reviewed on its own.
Perspective
The report covers the released feature scope of ThinkRail: workspaces cut as git worktrees, SPEC.md spec graphs beside the code, concurrent pi sessions and per-extension switches, addressed to engineers who edit repositories with coding agents day to day. Full extension support is named as the next step, at which point ThinkRail is to create the tools it needs from how a developer actually works.
Creating tools from how you work is not in place yet; it is the extension support named as the next step. MCP servers a repository defines stay off by default, calls the server does not mark read-only are confirmed in the chat first, and only tools trusted within a project can be allowed for good.