Two tools.
One workflow.
Move at terminal speed.
Why this feels different
The value is the workspace, not just the model.
fluidstate is useful because it turns agent work into something you can actually drive: multiple panes, real terminal output, built-in tooling, and clear moments where a human should step in.
"Ask the AI to fix the bug. Watch it work. Ask it to write tests. Watch it work again. One thing at a time, forever."
Result
- •You're the bottleneck
- •One context window, one task
- •Losing track of which tab is doing what
"Backend is editing. Another pane is running tests. One task is blocked on credentials. You can see the whole queue without losing the terminal."
Result
- •More than one branch of work can move at once
- •You intervene at the exact points where judgment matters
- •The tooling stays inside a keyboard-driven terminal workflow
Core workflow
How it actually works
The codebase itself points to the real advantages: real PTYs, flexible pane layouts, built-in navigation, and a workflow that keeps editing, diffing, and approvals in one place.
One binary. Done.
No Electron. No Node. No runtime.
fluidstate is a single Rust binary. Install with one command and run it in any directory. Works on macOS, Linux, and WSL2. Nothing to configure, nothing to update manually.
The layout expands with the work.
Multiple agent panes, each with a real terminal behind it.
The grid is not fixed to one canned demo state. Open focused panes, side-by-side sessions, or denser layouts as the workload grows. Every pane keeps its own context, terminal state, and scrollback.
Navigate, inspect, and edit without context switching.
Palette, file tree, editor, diff view, and approval flow in one shell.
The product is valuable because the surrounding tooling is already there. Open files, inspect diffs, search commands, switch focus, and handle permissions without dropping back into a pile of separate terminal tabs.
Use cases
Built for how engineering work fragments
Solo builders and teams both waste time in the same place: waiting for sequential AI loops to finish before the next branch of work can even start.
One Dev, Ten Agents
You're building alone but shipping like a team. Spin up agents for frontend, backend, tests, and infra—all running at once. Review and merge when they're done.
- →Task: “Build a Stripe integration with webhook handling, add a billing page, and write E2E tests.”
- →Agents: 3 agents fan out—one on the API, one on the UI, one on tests. All running simultaneously.
- →Result: “All 3 done in 12 minutes. Agent #2 needs your Stripe test key. The other two are ready to merge.”
Your Sprint, Parallelized
Turn a week-long sprint into an afternoon. Assign tickets to agents, monitor progress from a single board, and only step in when a human decision is required.
- →Monday: 15 tickets assigned to agents across 4 repos. Team grabs coffee.
- →By lunch: 11 PRs ready for review. 3 agents waiting on design decisions. 1 blocked on a flaky test.
- →End of day: Sprint shipped. Retro cancelled. Team wonders what to do with the rest of the week.
Tree — your repo as a live blueprint.
A graph-first desktop IDE that maps any repo into pages, endpoints, tables, hooks, and the connections between them — then runs agents (shell, claude, codex) alongside it. Every write on disk flashes the touched node in real time.
Unsigned build — on first launch, right-click Tree.app → Open to bypass Gatekeeper. Or install from source:
A standalone desktop app. Pair it with fluidstate in the terminal.
Install in one command
A pure terminal app with a tighter shell around it. No runtime, no Node, no Electron.
Then run fluidstate in any project directory.
Requires curl · macOS 12+
Contact
Your Codebase is Waiting.
Let's Ship It.
If the product already feels close, the landing page should close that last gap too. Reach out for access, feedback, or pilots.
© 2026 Fluid State. All rights reserved.