Fluid State is the product vision: agentic coding should move from intent to execution to system understanding without breaking context. v2 shapes the plan, fluidstate runs the agents, and Tree shows how the codebase changes.
The vision spans planning, execution, and architecture: v2 for the product layer, fluidstate for the terminal, and Tree for the codebase map.
Teams at
Future 500
Where the next era ships from
Not just today's Fortune 500 — the companies redefining how software gets built. Agentic workflows, parallel agents, and architectural clarity at scale.
Major corps
Used in production environments
Engineering teams run fluidstate in CI pipelines, internal tooling, and daily dev loops. Tree pilots sit alongside existing IDEs for large-repo refactors and onboarding.
Impact
Less waiting, more parallel work
Teams report cutting sequential AI loops from hours to minutes — with clearer visibility into what agents touched and where human judgment actually matters.
The system
Three products. One workflow.
Start with product intent in v2, execute through fluidstate, and watch architecture move in Tree. Each product stands on its own, but the full vision is the handoff between them.
fsfluidstate
CLITerminalAvailable
The agent workspace in your shell.
A single-binary CLI for running Claude Code, Codex, and shell sessions as parallel panes with real terminal control.
Parallel agent panes with real PTY output
Claude Code, Codex, and shell sessions side by side
Agentic work breaks when product intent, terminal execution, and codebase understanding live in separate places. Fluid State connects the handoff between all three.
Disconnected
“Write a spec in one tool. Run agents in another. Review the diff somewhere else. Reconstruct the architecture from memory.”
Result
•Product intent gets separated from implementation
•Parallel agents run without a shared map
•Review happens after context has already leaked away
v2 → fluidstate → Tree
“v2 turns the goal into scoped work. fluidstate runs the agents. Tree shows every code path the work touches as it lands.”
Result
•Intent, execution, and architecture stay linked
•Each product can stand alone or hand off to the next
•Human review happens with the original goal still visible
Workflow
Each product owns a clear step.
v2 shapes the work, fluidstate executes it, and Tree keeps the architecture visible. Use one surface or move through the full loop.
fsCLIfluidstate
01Install
One binary. Start in any repo.
Run the installer, open a project, and launch agent panes without adding a Node or Electron runtime to the workflow.
02Fan out
Split work across agents.
Give API, UI, tests, or review work to separate panes while keeping every session visible and interruptible.
03Land
Review the exact changes.
Stay in the shell for diffs, permission checks, command history, and the final judgment calls before code lands.
IDETree
01Open
Point Tree at the repo.
Tree scans the project and groups files into the architectural roles that matter for real product work.
02Watch
See the blast radius live.
As agents edit files, the graph flashes the touched nodes so you can tell which pages, routes, and schemas are moving.
03Steer
Keep context while agents run.
Use the map and panes together to decide where to intervene, where to review, and where the system is already coherent.
v2Planningv2
01Shape
Start with the product goal.
Capture the desired outcome, constraints, open questions, and acceptance checks before agents touch the repo.
02Scope
Break intent into runnable work.
Convert a broad ask into focused implementation, test, design, and review tracks that can run in parallel.
03Dispatch
Send the right work to the right surface.
Move execution into fluidstate for speed, Tree for architecture-heavy work, or both when the job needs multiple views.
Who it's for
Pick the surface by the job.
The main page is the vision. The individual pages are the details: planning, terminal execution, and architectural visibility.