Skip to content

DocsQuickstart

FutureX quickstart

Five steps from an empty terminal to a shipped change. Everything here runs the same way on Windows, macOS, and Linux — the one prerequisite is Node.js 20 or newer.

All documentation pages

1. Install the CLI

FutureX installs from the futurex-cli package and puts a single global binary on your PATH called futurex — that is the command you type from here on. Pick your package manager; the install line is the only thing that changes.

futurex-cli
npm i -g futurex-cli
# then: futurex login

Platform notes

  • WindowsPowerShell, Command Prompt, and Windows Terminal all run the same install line. If the shell cannot find futurex straight afterwards, open a new terminal so it picks up the global bin directory.
  • macOS · LinuxAny shell works. If a global install asks for elevated permissions, install Node through a version manager instead of a system package and run it again without sudo.
  • Node 20+Older runtimes are not supported. Check yours with node -v before you file a bug.

Confirm the install:

futurex --version
# prints the installed futurex-cli version

2. Sign in

Sign in once per machine. futurex login opens your browser with a one-time link code; confirm it in the portal and the terminal picks up the session.

terminal
futurex login

On a remote box, a container, or an SSH session where no browser can open, add --no-browser and FutureX prints the code instead — open the link on whatever machine has a browser.

futurex login --no-browser

Check the result with futurex whoami, which shows the backend and auth mode you are on. futurex logout clears the stored credentials again.

  • Non-interactiveFor CI and scripts, set FX_API_KEY in a .env file at the project root — it is loaded automatically when you run futurex, so no browser step is involved.

3. Index the repo

Change into your project and index it. Indexing is what makes answers retrieval-augmented: instead of you pasting files into a prompt, every turn pulls the most relevant chunks of your own code into context.

terminal
cd my-project
futurex index
# run it again after a large refactor

Two more indexes come with it. futurex symbols [name] queries the AST symbol index, and futurex imports [file] prints the import graph together with any broken imports it finds.

futurex symbols createUser
futurex imports src/app/page.tsx
  • Monoreposfuturex index scans from the workspace root, so one run at the top covers every package underneath it.
  • Turning it down--top-k n sets how many chunks are retrieved per turn (default 6). --no-context skips retrieval entirely for a single run.

4. Send your first prompt

Two ways in. Run futurex on its own to open the interactive REPL and stay in a conversation, or use futurex ask "<prompt>" for a one-shot answer that prints and exits — the form you want inside a script.

terminal
futurex
futurex ask "what does the auth middleware actually check?"

Every run defaults to the Fast module. Reach for --pro (or --tier fx-pro) on work that needs the extra depth, -r to raise reasoning effort, and --max for maximum reasoning. The four modules and what each one costs are laid out in Models.

  • Project memoryfuturex init creates futurex-memory.md at the repo root if it is missing. That file is how decisions survive between sessions — see Project memory.
  • Machine-readable--json makes ask and agent emit JSONL events on stdout and exit 0 or 1.

5. Ship a change

futurex build "<task>" is the whole loop in one command: it writes a plan — goal, value and budget, research, tooling, a task graph — and then executes against that plan. futurex plan stops after the plan if you want to read it first.

terminal
futurex build "add a health check route and a test for it"
# futurex plan runs the same first half and stops

Reads are automatic — FutureX opens your files without asking. Writes are not: the first time it wants to edit a file or run a shell command, it asks. Every mutating call resolves to one of four kinds — edit, create, delete, run — and you answer per kind.

Your answers are stored under your home directory, never inside the repository: a cloned repo must not be able to grant itself permissions. Allow and deny rules can live in either settings layer (~/.futurex/settings.json, then <repo>/.futurex/settings.json), and a deny is always checked first — so a repo can tighten your permissions but never widen them.

futurex permissions show
futurex permissions set edit allow
futurex permissions reset
  • VerificationAfter edits an agent run checks the project — typecheck, lint, test. --no-verify skips it; futurex verify runs the diagnostics on demand, and --list shows what it would run.
  • Picking it back upfuturex resume continues the latest session; --list and --session id choose another one.
  • Handle with care--auto-approve lets futurex agent run mutating tools without prompting. Keep it for throwaway repos.

Where to go next

That is install to shipped change. The reference pages go deeper on the parts you will touch every day.

  • CommandsEvery command and flag the CLI accepts, in one page — Commands.
  • ModelsPro, Fast, Light, and Eco: what each module is for and what it costs — Models.
  • PricingUsage-based, no per-seat subscription. The full terms live on Pricing.

ready when you are

You need an account before futurex login has anything to sign into. New accounts get the first month free.

Free first month · no card required · then $1 per 1M tokens on Fast, $2 on Pro