Viberails - Open-Source AI Agent Guardrail Tool

TL;DR

TL;DR: Viberails intercepts, audits, and validates AI agent tool calls before execution — acting as a security guardrail between AI coding assistants and their operations. Installs in minutes with a single shell command and adds under 50ms latency to tool calls.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: viberails.io ← verified via direct fetch
• Source repository: github.com/refractionPOINT/viberails ← README read in full
• License: Apache License 2.0 ← verified in README
• HN launch thread: news.ycombinator.com/item?id=44436137
• Docs: docs.viberails.io (coming soon per README)
• Built by Refraction Point, the team behind LimaCharlie
• Source last checked: 2026-07-06 (commit master branch)

What Is Viberails?

Viberails is an open-source security layer that intercepts tool calls from AI coding assistants before they execute. It sits between the AI agent and the actual operations — file edits, shell commands, API calls — and gives security teams and developers visibility and control over what AI tools are actually doing.

The project description on GitHub reads:

Enterprise-grade visibility and control for AI coding assistants

Viberails works as a proxy that AI coding tools route their operations through. When a tool call is made, Viberails can:

• Allow it to execute normally
• Block it and require human approval
• Log it for audit purposes without blocking

This is particularly relevant for teams using AI coding assistants in regulated environments or with access to sensitive infrastructure, where “hope nothing goes wrong” is not an acceptable security posture.

Supported AI Tools

Viberails supports a range of popular AI coding assistants. Per the README, the current supported tools are:

• Claude Code (Anthropic)
• Cursor
• Gemini CLI (Google)
• GitHub Copilot CLI
• OpenAI Codex CLI
• OpenCode
• OpenClaw (Refraction Point’s own agent framework)

Support for additional tools is actively being added.

Setup Workflow

Step 1: Install Viberails

macOS / Linux:

bash <(curl -fsSL https://get.viberails.io/install.sh)

Windows (PowerShell):

irm https://get.viberails.io/install.ps1 | iex

The install script downloads the binary for your OS and launches an interactive setup. If you prefer not to pipe curl to bash, you can manually download binaries from get.viberails.io:

| OS | Architectures | |---|---| | Linux | x64, ARM64 | | macOS | x64, ARM64 (Apple Silicon) | | Windows | x64, ARM64 |

Verify checksums via the release.json endpoint before running binaries if you want to be thorough.

Step 2: Create Your Team

After installation, the interactive setup will:

1. Select an OAuth provider (Google, Microsoft, or GitHub)
2. Enter your team name
3. Complete authentication in your browser

If you already use LimaCharlie, you can connect to an existing organization:

viberails init-team --existing-org <OID>

Step 3: Access the Dashboard

After setup, your team’s activity is visible at app.viberails.io. From there you can:

• Real-time Activity Feed — watch AI tool usage across your team live
• Policy Management — define rules for what tools can and cannot do
• Analytics and Reporting — understand AI tool usage patterns across the organization
• Alerts — get notified when sensitive operations occur
• Compliance Reports — generate audit reports with one click

Deeper Analysis

How It Works

Viberails operates as a proxy in the tool call chain. When an AI assistant attempts to perform an operation (read a file, run a shell command, make an API call), the request routes through Viberails first. Viberails then applies your configured policies before either passing the call through, blocking it, or flagging it for human review.

The README frames it starkly: the alternative is “hope nothing goes wrong.” Viberails replaces that hope with actual visibility and control.

Privacy Modes

Viberails offers three data handling modes so teams can choose how much information is sent to the cloud:

• Full audit — tool calls plus prompts are logged
• Tool authorization only — prompts are not logged, only tool call metadata
• Fully local — all processing stays on your machine, cloud features disabled

This gives security-conscious teams flexibility depending on whether they want complete audit trails or minimal data exposure.

Fail-Safe Design

The tool implements configurable fail-open or fail-closed behavior. Fail-closed means blocked operations stay blocked if something goes wrong; fail-open means operations proceed if the policy engine is unreachable. Teams can choose based on their risk tolerance.

Building from Source

Viberails is written in Rust. To build from source:

git clone https://github.com/refractionPOINT/viberails.git
cd viberails
cargo build --release
./target/release/viberails --help

The project enforces cargo clippy -- -D warnings (no warnings allowed) and all tests must pass before submission, enforced by pre-commit hooks.

The binary installs to ~/.local/bin/viberails. Add it to your PATH if needed:

export PATH="$HOME/.local/bin:$PATH"

Practical Evaluation Checklist

• [ ] Install completed in under 5 minutes
• [ ] OAuth team setup succeeded
• [ ] Dashboard accessible at app.viberails.io
• [ ] A test tool call appears in the real-time feed
• [ ] Policy rules can be configured (block/allow specific operations)
• [ ] Latency impact is under 50ms for a simple tool call
• [ ] Privacy mode switcher works (full audit / tool-only / local)
• [ ] Fail-open/fail-closed toggle is configurable
• [ ] Works with at least one of the supported AI coding tools

Security Notes

• Install verification: The install script can be reviewed at https://get.viberails.io/install.sh before piping curl to bash.
• Checksum verification: get.viberails.io/release.json provides checksums for all published binaries.
• Source is open: Apache 2.0 license, source on GitHub, contributions require passing clippy and all tests.
• Built by Refraction Point: The team behind LimaCharlie (an established EDR/XDR platform), giving this project credible security pedigree.
• Data exposure is opt-in: Three privacy modes let teams control how much data leaves their environment.
• No mention of SOC 2 or third-party security audits in the README — evaluate accordingly for enterprise procurement.

FAQ

Q: Does Viberails work with all AI coding assistants? A: No. The README lists seven specifically supported tools: Claude Code, Cursor, Gemini CLI, GitHub Copilot CLI, OpenAI Codex CLI, OpenCode, and OpenClaw. It does not support every AI tool. If your tool is not on the list, open a GitHub issue to request support.

Q: What is the latency overhead? A: The website states under 50ms latency for intercepted tool calls. Actual performance will depend on your network and configuration.

Q: Is there a cloud-only option? A: Yes. Team features (dashboard, policy management, analytics, alerts) require the cloud service. However, a fully local mode exists where all processing stays on your machine and no data is sent to Viberails’ servers.

Q: What is the pricing model? A: The website says “Get started for free, no credit card required” for individual developers. Enterprise pricing was not specified on the main page at time of writing.

Q: Can I block specific operations entirely? A: Yes. Viberails supports writing rules to block file deletions, restrict endpoint access, or require human approval for specific categories of operations.

Q: How is this different from just reviewing AI tool logs? A: Viberails enforces policies at execution time, not after. You can block or require approval for specific operations before they run, not just observe what already happened.

Q: Who builds Viberails? A: Viberails is built by Refraction Point, the same team behind LimaCharlie (a widely-used EDR/XDR platform). The project is Apache 2.0 licensed and open to contributions.

Conclusion

Viberails addresses a real gap in the AI coding assistant ecosystem: once you hand an AI agent access to your codebase and shell, you have limited visibility into what it actually does. Viberails makes that visible and controllable without slowing down developers significantly.

The install is trivial, the latency overhead is minimal, and the team behind it has security credibility. If you are deploying AI coding tools in a team environment — especially with sensitive code or infrastructure — a guardrail layer like Viberails is worth evaluating.

Project links:

• viberails.io — product page and dashboard
• github.com/refractionPOINT/viberails — source, issues, contributions
• app.viberails.io — team dashboard
• HN discussion — community discussion