# Guardrail by NeatProxy

Guardrail is a local spend firewall for AI coding agents. It records usage metadata and, on supported proxy integrations, checks budget policies before forwarding requests to providers.

- Findings24: <https://findings24.com/products/guardrail-by-neatproxy>
- Website: <https://www.neatproxy.com>
- Pricing: Freemium
- Categories: AI Agent Infrastructure, AI Gateway And Routing

## What is Guardrail by NeatProxy?

Guardrail is a local spend firewall for AI coding agents. It runs on the user’s machine and places supported tools in front of their providers through a local proxy at `127.0.0.1:4000`.

For supported proxy integrations, it records usage metadata and evaluates spend, request, token, and rate policies before a network call. Requests that violate a configured policy are refused locally rather than forwarded. Guardrail does not store prompts, responses, or code in its usage records.

Support depends on the tool and login mode: Claude Code and Codex with an OpenAI API key support pre-send policy checks, while Codex with a ChatGPT subscription provides telemetry only. Cursor support is planned.

## What can Guardrail by NeatProxy do?

### Pre-send budget enforcement

Evaluates spend, request, token, and requests-per-minute policies before a supported proxied request reaches the provider. Over-policy requests are stopped locally.

### Local usage metering

Writes model, token, cache, estimated-cost, timing, and status metadata to a local SQLite file without storing prompts, responses, or code.

### Local dashboard and history

Shows spend, sessions, and cost details locally. The Free plan includes seven days of history; Pro and Team include 90 days.

### Tool-specific integrations

Connects Claude Code and Codex through command-line configuration, with different enforcement behavior for Codex API-key and ChatGPT-subscription modes.

### Cloud and team controls

Pro adds a hosted dashboard and sync across machines. Team adds an admin console, per-member spend, and shared project caps for supported proxy integrations.

## Use Cases

### Limit unattended coding-agent runs

Set a session or project policy before an agent runs unattended, so supported API requests that exceed the policy are stopped locally instead of forwarded.

### Inspect hidden request costs

Review token, cache, reasoning-token, and estimated-cost metadata to understand what contributed to a coding session’s spend.

### Track usage across developer machines

Use Pro cloud sync to review supported local usage from multiple machines while keeping prompts, responses, and code out of synced data.

### Manage shared team budgets

Use Team’s admin console, per-member spend, and shared project caps to monitor and control supported proxy usage across a group.

## Frequently Asked Questions

### Does Guardrail block every AI coding-agent request?

No. Blocking requires a supported proxy integration and depends on the tool’s login mode. Claude Code and Codex with an OpenAI API key support pre-send policy checks. Codex with a ChatGPT subscription is telemetry-only, and Cursor support is planned.

### How is Guardrail installed and started?

The installer downloads and verifies a platform release and places the \`guardrail\` binary on the PATH. Run \`guardrail start\`, connect a supported tool with \`guardrail connect claude-code\` or \`guardrail connect codex\`, and configure a budget in the dashboard. The daemon binds to \`127.0.0.1:4000\`. Windows installation currently uses WSL2 while a native build is in progress.

### What data does Guardrail store?

Local records contain metadata such as model, token counts, cache activity, estimated cost, timings, and status in SQLite. Prompts, responses, and code are not stored. Cloud sync sends metadata only; personal sync is opt-in, while Team machines sync by default on supported versions and can opt out.

### What happens when a request exceeds a policy?

For supported proxy integrations, Guardrail stops the request before the network call and returns a clean 429 or policy error to the tool. How the tool retries or recovers depends on that tool.

### What is included in the Free plan?

Free does not expire and requires no email or credit card. It includes the local firewall, budget policies, local dashboard, and seven days of history. Pro costs $15 per month or $144 per year and adds cloud features; Team is offered as flat bundles for up to five or 10 seats.

## Quick Facts

- **Product type**: Developer tool / local AI spend firewall
- **Primary users**: Developers and teams running AI coding agents
- **Supported blocking modes**: Claude Code; Codex with an OpenAI API key
- **Telemetry-only mode**: Codex with a ChatGPT subscription
- **Local endpoint**: 127.0.0.1:4000
- **Free plan**: Local firewall, policies, dashboard, and seven days of history