@flumecode/terminal

A terminal-based coding agent that manages your work through the FlumeCode platform. It routes every turn, gates every tool call, and auto-manages task lifecycle — creating tasks before code edits, marking them done on merge, and canceling them when you stop.

The model backend is Fireworks AI (default model: GLM 5.2, 1M-token context window). You'll need a Fireworks API key to use it.

For developers: architecture, dev setup, and internals are in TECHNICAL.md.

Install

npm install -g @flumecode/terminal

Or run without installing:

npx @flumecode/terminal

Requires Node.js 20 or later.

Quick start

1. Get a Fireworks API key at https://fireworks.ai/account/api-keys.

2. Set it as an environment variable:

   export FIREWORKS_API_KEY="fw_..."

3. Run the terminal:

   flumecode-terminal

   The first time you run it, the terminal opens guided setup — sign in or create an account, connect your model API key, and confirm your model. Once you're signed in it boots straight into the prompt on every later run; just start typing a request and press Enter to send.

Guided setup (/onboard)

Guided setup runs automatically until you're signed in, and you can re-run it any time with /onboard. It walks you through signing in (or creating an account), connecting your model API key, and confirming the model — step by step. You never leave the terminal.

Switching accounts

Sign in or switch accounts mid-session with /login (or create one with /signup). The terminal re-boots into the new account in place — it doesn't quit.

Local mode (developers): running from a monorepo checkout (or with FLUMECODE_DEV=1) boots against local dev servers backed by an embedded database, so no account is needed. Local mode is a development convenience and is not available in the published package.

Slash commands

Type any of these at the prompt (they're also shown in a live preview menu as you type /):

| Command | Description | | --- | --- | | /onboard | Guided setup: sign in, connect your model, and confirm the model id | | /connect [key] | Set the AI provider API key (masked prompt if no key given) | | /model [id] | Show the active model, or switch to <id> for the next turn | | /compact | Summarize the conversation into a recap and drop the full transcript | | /clear | Start a fresh conversation and wipe the on-screen history | | /login | Sign into a hosted account and re-boot the session | | /exit | Quit the terminal |

/compact vs /clear

• /compact — summarizes the running transcript into a recap, then replaces it in-place so the context window doesn't fill up. Preserves the task thread.
• /clear — starts completely fresh: drops the transcript, the task thread, and wipes the screen. Use when you want a clean slate.

Auto-compaction fires at 90% of the context window after each turn, so you rarely need to run /compact by hand.

Key bindings

| Key | Action | | --- | --- | | Enter | Send your message | | Shift+Enter | Insert a new line | | Esc | Interrupt the model mid-turn | | Ctrl-C (twice) | Exit the terminal | | Ctrl-D | Exit the terminal |

Environment variables

| Variable | Purpose | | --- | --- | | FIREWORKS_API_KEY | API key for Fireworks AI (required to use the model) | | FLUMECODE_MODEL | Override the default model id (e.g. accounts/fireworks/models/glm-5p2) |

Config files

Settings are stored in ~/.flumecode/:

| File | Contents | | --- | --- | | ~/.flumecode/provider.json | Provider choice, API key, and model (permissions: 600) | | ~/.flumecode/config.json | Hosted account login token |