Infiniax Code

The official autonomous coding agent from InfiniaxAI

Infiniax Code is an autonomous AI coding assistant that lives in your terminal. Point it at any project, give it a task in plain English, and it plans the work, edits files, runs commands, reviews its own changes, and rolls back on failure. Every action is shown live with its real cost in cents, and you stay in control with per-action permissions.

cd ~/your-project
npx infiniax-code

What's new in v0.3

A focused polish pass on access, speed, clarity, and trust. Highlights:

Open to free users — full access, every model

• Free plans now welcome, with every model unlocked. The CLI used to require a paid Starter/Pro/Unlimited plan. It no longer does. Every signed-in InfiniaxAI account, including the free plan, can run infiniax-code and use every available model — Claude, GPT-5.5, Haiku 4.5, the entire lineup. No model is locked behind a tier.
• Credit balance is the only limit. Free users spend from their monthly free allotment. If the balance runs out mid-task — even in the middle of an edit or a long multi-step run — the agent stops cleanly right there, surfaces an OUT_OF_CREDITS message, leaves any in-flight changes intact, and waits. Top up, upgrade, or wait for your monthly reset, then re-run and continue.

Speed & cost

• /spark fast mode — flip the agent into a Haiku 4.5 sprint that skips the planner, tightens the tool budget, and answers small tasks in one or two calls. Toggle on with /spark, off the same way.
• Trivial question fast path — short questions like what does X do? or explain this file skip the planner entirely and answer directly from a single think call.
• Grounding cache — the project scan is cached for the session and only invalidated when a tool actually writes a file, so the agent stops re-walking your repo between actions.
• Cumulative working timer — Working (1m 42s · esc to interrupt) now counts the whole turn instead of resetting to 0s every action, so you finally see real elapsed time.

Reliability

• Patcher hardening — bumped the unified-diff fuzz factor and added a whitespace-agnostic block-match fallback. Edits that previously failed with Patch could not be applied because of indentation drift or reformatted context now apply cleanly.
• Empty-response model fallback — when an upstream model returns nothing or errors, the runtime now transparently retries on a known-good fallback model from the same family.
• Streaming finalization fix — final messages and the streaming bubble swap in the same render commit so the UI no longer flickers at the end of a response.

Interface

• True token streaming with heavier bold — markdown streams progressively as the model emits it, and **bold** renders with full-weight ANSI 1m glyphs.
• Single clear cost line — the old box panel is gone. Every run ends with one bright line: Worked 1m 24s · $0.28 · 7 actions · 3 files. Per-action deltas remain inline beside each step.
• Animated splash + outlined sign-in — a calmer, cinematic monochrome intro with a one-time browser-based login flow.
• Inline placeholder hint — /spark for fast mode, /compact to compact context, /btw for instant questions — all shown inline under the prompt without polluting your scrollback.
• No emojis anywhere — strict monochrome, status glyphs only (· ✓ ✗).

Slash commands

• /spark toggle, /compact to shrink context mid-conversation, /thinking to set reasoning effort (Low → Max), and a refined /model picker. /models was removed in favour of the single /model command.

Install

The easiest way to run Infiniax Code is with npx. No install, no permission setup:

cd ~/your-project
npx infiniax-code

Requires Node.js 18 or newer. Works on macOS, Linux, and Windows.

Global install (optional)

If you'd rather have the command on your PATH permanently:

npm install -g infiniax-code
infiniax-code

EACCES on macOS or Linux? npm install -g writes to /usr/local, which is owned by root on many systems. Pick one:

• npx infiniax-code (recommended, no install needed)
• sudo npm install -g infiniax-code
• Install Node via nvm so npm installs into a folder you own.

Get it running with your InfiniaxAI account

Infiniax Code uses your existing InfiniaxAI account for authentication and billing. There are no separate API keys to manage.

Free users are welcome — full access, every model. Every InfiniaxAI account, including the free plan, can run real tasks in the CLI and use every available model (Claude, GPT-5.5, Haiku 4.5, the full lineup — no model is locked behind a tier). The only limit is your credit balance: free users spend from their monthly free allotment, paid plans (Starter, Pro, Unlimited, Business) spend from their larger included balance plus any additional credits they buy. If you run out mid-task the CLI stops cleanly right there — even in the middle of an edit or a long multi-step run — surfaces a clear OUT_OF_CREDITS message, leaves any in-flight changes where they are, and waits. Top up, upgrade, or wait for your monthly reset, then re-run the task and pick up from where it stopped.

1. Create an account at infiniax.ai if you don't already have one. Free plan is fine.
2. Run the CLI in any project directory:

   cd ~/your-project
   infiniax-code

3. Sign in when prompted. The CLI opens your browser to infiniax.ai for a one-time login. Credentials are saved locally so you only do this once per machine.
4. Start working. Type a task and press Enter.

Every action is billed against your InfiniaxAI credit balance at the same per-token rates as the web app, plus a small CLI multiplier that covers the agent loop's amplified token usage. The exact cost of each action is printed inline, and a single cumulative cost line ends every run.

Tip for free users: turn on /spark mode (fast mode, Haiku 4.5, no planner) and start with small tasks. The free monthly credit pool stretches noticeably further when the agent isn't running multi-step plans on every prompt.

Using it

Just describe what you want in plain English. The agent figures out the rest.

infiniax-code

> add rate limiting to the /api/login endpoint
> explain how the auth flow works in this codebase
> the dashboard chart is broken on mobile, fix it
> write tests for the payment service

Slash commands

Type / at any prompt to see a live menu. As you type, the list narrows.

| Command | What it does | |---|---| | /spark | Toggle fast mode — Haiku 4.5, no plan, tight tool budget | | /model | Pick which AI model powers the agent | | /thinking | Set reasoning effort: Low, Medium, High, xHigh, or Max | | /permissions | Choose Ask, Edit Automatically, or Full Access | | /compact | Compact the context window mid-conversation | | /btw <question> | Ask a quick side question while the agent is working | | /update | Check for and install the latest version | | /help | Show all commands |

Keyboard shortcuts

| Key | Action | |---|---| | Tab | Autocomplete a slash command | | Esc | Interrupt the agent mid-run (it stops cleanly and reports as Interrupted) | | Ctrl+C | Exit the CLI |

What the agent can actually do

• Read, edit, and create files anywhere in your project
• Search across the codebase
• Run shell commands
• Install packages
• Query your database
• Take screenshots and verify the result of UI changes
• Spawn subagent reviewers for deep code review
• Roll back its own changes when something fails validation

You stay in control of all of it through the permission modes.

Links

• Web app: infiniax.ai
• Account & billing: infiniax.ai/settings
• Issues & feedback: Reply to any email from us, or hit /btw inside the CLI.

License

MIT — built and maintained by InfiniaxAI.