Skip to main content
Pre-release

This release is a pre-release and may not be stable for production use.

Neuron

Neuron by Zero-X is an open-source AI coding agent that runs in your terminal. It reads your project, edits files and runs commands, and asks before it does anything your permission mode does not allow. It is a single native binary written in Rust. New to coding agents? Read how an AI coding agent works.

By default Neuron talks to models through the Zero-X gateway. You sign in once with neuron auth login; the provider API keys stay on the gateway and never reach your machine. You can also use your own provider key instead.

7.0.0-alpha.3 is Neuron's third alpha, built and published by this repository's release pipeline. Its known limitations are listed in the changelog. It is a GitHub prerelease, so the install scripts need the versioned commands below. The npm 6.2.0 and PyPI 6.2.4 (and earlier) packages were built outside the pipeline and are not supported.

Install

Every method installs the same neuron binary from GitHub Releases and checks its SHA-256 checksum before using it.

A prerelease is never the "latest" GitHub release, so the install script commands below name the alpha's version. Without one, the install scripts look up the latest release, and there is none until the first stable release is out.

macOS and Linux

curl -fsSL https://github.com/RAHUL-DevelopeRR/neuron-v2/releases/download/v7.0.0-alpha.3/install.sh | sh -s -- --version 7.0.0-alpha.3

The script installs to ~/.local/bin and tells you if that directory is not on your PATH. Pass options after sh -s --, for example sh -s -- --version 7.0.0-alpha.3 --install-dir "$HOME/bin", or set NEURON_VERSION and NEURON_INSTALL_DIR.

Windows (PowerShell 5.1 or 7)

& ([scriptblock]::Create((irm https://github.com/RAHUL-DevelopeRR/neuron-v2/releases/download/v7.0.0-alpha.3/install.ps1))) -Version 7.0.0-alpha.3

The script installs to %LOCALAPPDATA%\Programs\Neuron\bin and only changes your PATH if you ask it to:

& ([scriptblock]::Create((irm https://github.com/RAHUL-DevelopeRR/neuron-v2/releases/download/v7.0.0-alpha.3/install.ps1))) -Version 7.0.0-alpha.3 -AddToPath

npm (Node.js 22.14 or later)

npm install --global @zero-x.live/neuron

pip (Python 3.9 or later). The alpha is a pre-release on PyPI, so pip needs --pre:

pip install --pre neuroncli

The registry packages are published after the GitHub release, so check them first: npm view @zero-x.live/neuron version should print 7.0.0-alpha.3, and pip index versions neuroncli --pre should list 7.0.0a3. If they still show 6.2.0 (npm) or 6.2.4 (PyPI), those are old, unsupported builds: use the install scripts above instead.

The npm and pip packages are small launchers that download the binaries for the package's exact version and verify their checksums. npm downloads them during npm install, without showing progress: neuron (about 17 MB) and neuron-tui (about 110 MB) go into the package's bin/native folder. pip downloads them on first use instead: neuron the first time you run it, and neuron-tui the first time the full-screen UI opens; npm does the same when the network failed during the install. pip keeps them in a per-version cache (~/.cache/neuroncli on Linux, ~/Library/Caches/neuroncli on macOS, %LOCALAPPDATA%\neuroncli on Windows). npm uses a binary from that cache when its bin/native folder has none, and downloads into the cache only when that folder is read-only. After that, Neuron starts without network access. Ctrl+C goes to Neuron, and the launcher exits with Neuron's exit code.

With the installers or pip, the two default MCP servers (CodebaseMap, which /mcp lists as codebase-memory, and Playwright) start through npx -y, so they need Node.js; the npm package installs them with Neuron.

On a terminal, neuron with no prompt opens the full-screen UI, neuron-tui. The installers put it beside neuron, and the launchers fetch it the same way, verified by checksum. neuron --no-tui (or NEURON_NO_TUI=1) keeps the line REPL. Without the UI, for example after a failed download, neuron runs the line REPL as before. Prompts (neuron "fix the test", neuron -p ...) and subcommands never start it.

From source

git clone https://github.com/RAHUL-DevelopeRR/neuron-v2.git
cd neuron-v2/rust
cargo build --release -p rusty-claude-cli
./target/release/neuron --version

Supported platforms

Platform Release assets
Linux x64, glibc 2.35 or later neuron-linux-x64, neuron-tui-linux-x64
Linux arm64, glibc 2.35 or later neuron-linux-arm64, neuron-tui-linux-arm64
macOS, Apple silicon neuron-macos-arm64, neuron-tui-macos-arm64
macOS, Intel neuron-macos-x64, neuron-tui-macos-x64
Windows x64 neuron-windows-x64.exe, neuron-tui-windows-x64.exe (Windows on Arm runs them under emulation)

glibc 2.35 or later means Ubuntu 22.04, Debian 12, Fedora 36 or newer. On older distributions (for example Ubuntu 20.04, Debian 11 or RHEL 9) the binary may not start; the installers check that before they install anything, and you can build from source instead. musl-based Linux such as Alpine is not supported. The binaries are not code-signed or notarized. Each release lists SHA-256 checksums (SHA256SUMS, plus a .sha256 file per binary) and GitHub build provenance, which you can check with gh attestation verify neuron-linux-x64 --repo RAHUL-DevelopeRR/neuron-v2.

Sign in

neuron login          # shows a code; approve it on neuron.zero-x.live/device
neuron auth status
neuron logout         # revokes the session on the gateway and deletes the local copy

Login prints a link and a short code and opens the link in your browser when it can. Approve the code there (or on any other device, such as your phone) and Neuron continues on its own; no local port is opened, so it also works over SSH. It stores the session in the operating system's credential store on macOS and Windows. On Linux (or when that store is unavailable) it goes to ~/.neuroncli/credentials.json, readable only by you; protect that directory like a password file. The security policy explains what to do if it leaks. On CI and machines without a browser, create a personal API key (nzx_…) in the Zero-X dashboard and pass it with neuron auth login --with-token or the NEURON_TOKEN environment variable. See Sign in in the usage guide.

Plans and limits

Free. No card needed. Free-plan requests run on free tiers of upstream providers: Groq, the Gemini API free tier, Cloudflare Workers AI, NVIDIA's API catalog and OpenRouter's free models. The limits are real and you will run into them:

Free plan Limit
Requests per day 100
Tokens per day 500,000
Requests per minute 10
Concurrent requests 2
  • Everyone on the free plan shares the same upstream capacity. When a shared pool runs out, requests fail with capacity_exhausted and a time to retry.
  • Some free models have small per-minute token budgets, so a long prompt may fit on only some models, or on none.
  • The Gemini free tier is not offered to callers in the EEA, the UK or Switzerland.
  • Which models are free depends on what the providers offer, and can change at any time. /model lists what is available to you now; outside a session, curl https://zero-x.live/v1/models shows the current catalog.

Pro and Max. Paid plans raise the daily and per-minute limits and add paid upstream capacity, billed against monthly credits included with the plan; credit top-ups are also sold. Prices and current limits are at https://zero-x.live/pricing. In Neuron, /upgrade opens that page, and neuron usage (or /usage) shows your plan, today's usage and your credit balance.

Bring your own key

You can bypass the gateway with a provider account of your own: for example OPENAI_API_KEY (plus OPENAI_BASE_URL for any OpenAI-compatible endpoint, including a local model server), ANTHROPIC_API_KEY, XAI_API_KEY, Azure AI Foundry or OPENROUTER_API_KEY. The gateway offers two of OpenRouter's free models; your own key gives you the rest of OpenRouter. While you are signed in to Zero-X the gateway stays the default; a provider key is used when you select that provider (--provider, NEURON_PROVIDER or the provider setting) or when you are not signed in. Neuron prints which provider and credential it uses when it starts. See Bring your own key.

Privacy

What you send in a session (prompts, the contents of files Neuron reads, command output) is processed by third-party model providers: the upstream provider the gateway routes to (Groq, Google, Cloudflare, NVIDIA, OpenRouter or a paid upstream), or the provider whose key you configured. Their terms apply. Free tiers can have different data-use terms from paid ones; Google, for example, may use Gemini API free-tier content to improve its products. Paid Zero-X plans also use the free pools first and fall back to paid capacity, so if your code must stay under one provider's terms, use your own key for that provider.

The gateway forwards each request to the upstream provider and records usage metadata (model, token counts, latency, status) for quotas and billing. Neuron sends no analytics of its own. Tools you allow, such as shell commands, web fetches and MCP servers, can contact other services.

Using Neuron

cd your-project
neuron                                   # interactive session
neuron prompt "explain src/main.rs"      # one prompt, then exit
neuron --resume latest prompt "and the tests?"   # one more prompt in the newest session
neuron --permission-mode read-only       # read and search only

Neuron asks you to trust a directory before working in it. The default permission mode, workspace-write, lets it read and edit files in the workspace and run shell commands that only read; it asks before any other command, before fetching from a new web domain and before other tools that need full access. The usage guide covers commands, slash commands, permission modes, models, sessions and configuration.

Scripts and CI

neuron --output-format json prompt "run the tests and summarise the failures"
neuron --resume latest prompt "now fix the first failure"

Launch flags go before prompt (or -p). Everything after -p is the prompt, so a flag such as --output-format there is refused rather than sent to the model. Without a terminal on standard input nobody can approve a tool call, so a call that would ask is denied instead, the model is told it did not run, and Neuron exits with code 2 after the answer. Each denial names the permissions.allow rules that would let that call run, when any would, and the last warning lists them all.

Allow what your script needs in your user settings, ~/.neuron/settings.json. A script never gets the trust question, so the workspace's own settings, allow rules included, are ignored until you trust it.

{
  "permissions": {
    "defaultMode": "workspace-write",
    "allow": ["bash(npm:*)", "PowerShell(npm:*)"]
  }
}

bash(npm:*) lets a shell command run when every part of it that needs approval is an npm command (npm install && npm run build). Rules count together: with bash(cargo:*) as well, cargo test && npm run build runs too. No prefix rule covers a command that writes a file through a redirection (npm test > out.txt), sets an environment variable first (CI=true npm test) or uses a substitution; a rule for the exact command does ("bash(npm test > out.txt)"). "WebSearch" allows web searches. Use --permission-mode danger-full-access only in a disposable environment such as a container: it lets every tool call run.

Each neuron prompt run starts a new session; neuron --resume latest prompt "..." sends the newest session another prompt. The JSON answer's session_id is what --resume takes, so a script can continue its own session by id. One run at a time can continue a session: a second one is refused as "in use" rather than overwriting the first one's turns.

Documentation

The parity, roadmap and architecture documents describe the upstream project Neuron grew out of and may still use its claw command names.

License

MIT; see LICENSE. Neuron is derived from Claw Code, also MIT. The full-screen UI (ui/neuron-tui, the neuron-tui binaries) is derived from Freebuff and is under the Apache License 2.0. NOTICE records both lineages and the bundled third-party skills, and every release attaches THIRD_PARTY_NOTICES.md with the licenses of the Rust crates compiled into neuron and of the Bun runtime and npm packages compiled into neuron-tui. Neuron and Zero-X are not affiliated with Anthropic, OpenAI, Google, Groq, Cloudflare or any other model provider.

Metadata

Release files for neuroncli 7.0.0a3

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for neuroncli 7.0.0a3
File Size Uploaded Provenance Actions
neuroncli-7.0.0a3.tar.gz 44.0 kB With provenance Details for neuroncli-7.0.0a3.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for neuroncli 7.0.0a3
File Interpreter ABI Platform Actions
neuroncli-7.0.0a3-py3-none-any.whl Python 3 none any Details for neuroncli-7.0.0a3-py3-none-any.whl

Total release size: 59.6 kB

Release files / neuroncli-7.0.0a3.tar.gz

Download URL neuroncli-7.0.0a3.tar.gz
Size 44.0 kB
Tags Source
SHA-256 checksum
How to use checksums
b38a02d35a9cb796699f394cfae0c4d86dbcd18edecb8644f4cf5d4952a282af
BLAKE2b-256 checksum
How to use checksums
4fbc6947e59cf9dc3b7dfc1e7fc042f0f825bb39f8d1dd6fc95b0343a94e5719
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 3, 2026.

Transparency log

Release files / neuroncli-7.0.0a3-py3-none-any.whl

Download URL neuroncli-7.0.0a3-py3-none-any.whl
Size 15.6 kB
Tags Python 3
SHA-256 checksum
How to use checksums
77df9bc16b4371eb346e0603070b1cc07b383fb3ff549a123f0b52859bcaa1cb
BLAKE2b-256 checksum
How to use checksums
f70a34b19bb53def1f63fe1f824a58d9055cd8c1144d960eec5ed9f692642d02
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 3, 2026.

Transparency log

Security for neuroncli 7.0.0a3

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for neuroncli-7.0.0a3.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 3, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page