futurex-cli

FutureX — the terminal-native AI coding agent from FIM.

Describe a change in plain English. FutureX reads your repo, plans the work, writes the code, runs your checks, and ships a preview — without you leaving the shell. Streaming REPL plus an agentic tool loop with permission gates, on Windows, macOS, and Linux.

• Website: https://futureim.org
• GitHub: https://github.com/FutureIM
• Requires: Node.js 20+

Install

npm i -g futurex-cli      # pnpm add -g futurex-cli · yarn global add futurex-cli · bun add -g futurex-cli
futurex login             # sign in via a browser link code
futurex                   # interactive REPL

New accounts get a free launch month. After that FutureX is usage-based — no per-seat subscription. See pricing.

The published binary is a single bundled file, so a global install pulls only the runtime dependencies.

Commands

futurex                     # interactive REPL
futurex ask "<prompt>"      # one-shot answer, then exit
futurex agent "<task>"      # agentic loop: reads run automatically, writes are permission-gated
futurex plan "<task>"       # goal + value/budget + research + tooling + task DAG
futurex build "<task>"      # plan, then execute against it until the goal is met
futurex resume              # resume your latest session (--list, --session <id>)

Repo intelligence:

futurex index               # index the repo for retrieval-augmented answers
futurex symbols [name]      # query the AST symbol index ("is <name> defined?")
futurex imports [file]      # import graph + broken imports
futurex init                # create futurex-memory.md (project memory)
futurex compact             # bound futurex-memory.md size (regenerate + archive)

Workflow and account:

futurex skills [request]    # list skills, or route a request to the matching ones
futurex verify [--list]     # project diagnostics (typecheck / lint / test / build)
futurex test                # run the project test script
futurex permissions         # per-project consent: show | set <kind> <state> | reset
futurex login / logout / whoami
futurex help

Options

| Flag | Meaning | | --- | --- | | --tier <module> | fx-pro · fx-fast · fx-mini · fx-eco | | --effort <level> | off · high · max (enforced against the module) | | -r, --reason | more reasoning effort (high) | | --max | maximum reasoning (implies fx-pro) | | --pro | use the fx-pro module | | --max-tokens <n> | cap output tokens | | --no-memory / --no-repo / --no-context | drop project memory / repo card / retrieved chunks | | --top-k <n> | chunks retrieved per turn (default 6) | | --auto-approve | (agent) run mutating tools without prompting — dangerous | | --no-verify | (agent) skip the automatic typecheck/lint/test pass after edits | | --max-steps <n> | (agent) max model/tool rounds (default 32) | | --json, --output-format <f> | machine mode: JSONL events on stdout, exit 0/1 | | --value <tier> | (plan/build) low · standard · high · critical | | --budget <n> | (plan/build) cap the build token budget | | --restart | (plan/build) ignore checkpoints and re-plan from scratch | | -v, --version | print version |

Note: --tier / --effort apply to ask and chat; the agent executor runs fast by design.

Permissions

Reads (read_file, list_dir, grep_search) run automatically. Writes and shell execution are gated: FutureX asks before it changes anything, and your answers are remembered per project. Inspect or reset them with futurex permissions.

Configuration

Settings are layered — user first, then project:

~/.futurex/settings.json
<repo>/.futurex/settings.json

They carry permissions.allow / permissions.deny rules, hooks, defaultTier, verify, and mcpServers.

| Variable | Purpose | | --- | --- | | FX_API_KEY | API key for non-interactive use, auto-loaded from your .env | | FUTUREX_API_URL | point the CLI at a different FutureX endpoint |

Links

• Product and pricing — https://futureim.org
• GitHub — https://github.com/FutureIM
• Support — [email protected]

FutureX is the coding engine of FIM (Future Information Mastery). Prompt. Build. Ship.