Install once. Code anywhere.
Use LayerFlow in your browser with zero setup — or install the terminal CLI and code from anywhere. Both share the same sessions, context and cost ledger.
Install the terminal CLI
One command, nothing else. No API keys, no config files, no account needed to try it.
macOS / Linux
$curl -fsSL https://layerflow.dev/install | bashWindows (PowerShell)
$powershell -ExecutionPolicy Bypass -c "irm https://layerflow.dev/install.ps1 | iex"Runs in PowerShell or Windows Terminal. Works in Git Bash / WSL too — just use the macOS/Linux command.
verify
$lf version→ lf 0.2.6 — the latest release
Quick start — terminal
Write plain English, click Improve in the web app, or just run it — LayerFlow sharpens the prompt, picks a model, checks cost, and runs agents for you.
$lfLaunch the full-screen terminal UI — home, streaming chat, sessions, model switcher and command palette.
$lf run "build a landing page"Run a single task with live step streaming and tool approvals — prompt, cost and model handled for you.
$lf chat "explain the auth flow"Start a streaming chat session with the LayerFlow gateway. Continue the same session in the web app.
$lf sessionsList persisted sessions for the current project and reopen any past one with `--id`.
$lf syncPush/pull your sessions, context and cost ledger between the terminal and the cloud.
$lf cost --projectShow token and dollar usage from the local store — keep an eye on what each project is spending.
Use it in the browser — zero install
Everything the CLI does, without installing anything:
Coding Workspace
Plain English in, working code out. Improve, then run with agents.
Improve in Chat
One click in any chat: your rough prompt becomes a sharp, low-token prompt — scored 0–100.
Build Agents
Create your own specialist agents with custom prompts and models, then run them from the web.
Rescue Chat
Paste a dead AI chat and get a clean prompt, compressed context and Continue Pack.
BYOK
Bring your own API keys. Pay providers directly — never resold credits.
Cost Check
Dollar estimates before every run, plus the cheapest good-enough model.
How it works
One session format everywhere. Web, terminal CLI and agents all write to the same session store, so nothing gets lost when you switch surfaces.
Prompt Improver
Plain English in. Clarity, context, constraints and output format added — scored 0–100 before it runs.
Agents
Implement, review and test agents run with tool calls: read/edit files, run commands, check diffs.
Context & sessions
Every run saves an AI summary, a prompt and a Continue Pack. Resume from any surface.