Guardrail by NeatProxy

Stop AI agents
before they overspend.

Guardrail is a local proxy for Claude Code and Codex. It tracks every call as it happens and blocks the request before it leaves your machine when a budget is hit. No cloud. No API keys. No prompts stored.

Claude Code & Codex · Local-first · Metadata only

guardrail · localhost:4000
The Guardrail dashboard: 353 requests and $3.78 estimated cost over seven days, one request blocked before it reached the provider, and a daily cost chart split by Claude Code and Codex.

The dashboard runs on your machine, not ours.

The problem

Agents spend money faster than anyone can watch.

A coding agent runs unattended for minutes at a time, and a single retry loop can burn more than a month of subscription before anyone looks. The tools that exist today tell you afterwards.

01

Bills arrive late

Provider billing is a monthly rear-view mirror. By the time a number looks wrong, the money is already spent.

02

Counters only count

A token counter watches the meter run. It has no opinion about when to stop, and no way to act on one.

03

Alerts are not brakes

A threshold alert fires after the spend it is warning you about. A budget you cannot enforce is a preference.

How it works

One command in front of your agent.

Guardrail is a local proxy. It sits between your coding tool and the provider, relays your existing login, and reads only the metadata as requests pass through — so it can show you the spend and stop it before it leaves your machine.

1

Connect, keyless

One command points your tool at Guardrail and relays your own Claude Code or Codex login. No API key changes hands — nothing to vault or rotate.

2

It reads the metadata

Every call flows through localhost:4000 byte-for-byte. Guardrail parses model, tokens, cache and cost — never the prompt, the response, or your code.

3

See it, then cap it

A live local dashboard shows spend, sessions and hidden cost. Set a budget and an over-budget call is blocked before it ever reaches the provider.

Claude Code · CodexGuardrail · :4000Anthropic · OpenAI
SQLite · metadata only

The request body is forwarded untouched. Only model, token counts and an estimated cost are written down — locally, bound to 127.0.0.1.

What Guardrail controls

Spend control, not another dashboard to babysit.

Six things it does on your machine, with no account and no cloud. See how each one works.

Control

Budgets that block

Caps on dollars, requests, tokens or requests-per-minute. Over budget, the call is never sent, so it costs nothing.

Visibility

Hidden cost, surfaced

The cache-write tax on a first turn and the reasoning tokens you never see, attributed per request.

Privacy

Keyless and local

It relays your own login, so no key changes hands, and keeps metadata only in a file on your machine.

Coverage

Claude Code, Codex, agents

One proxy for both coding tools, plus any agent framework that speaks the OpenAI Chat Completions API.

Forecast

Know the month before it lands

Spend pace and a projected month-end total from the sessions you have already run.

Flow

Pause without restarting

Flip tracking off and back on from the dashboard. Your tool keeps working throughout.

Why Guardrail is different

Prevention, not reporting.

Most tools in this space observe. Guardrail sits in the request path, which is the only place a budget can actually be enforced.

Provider dashboards

Show you the bill

Accurate and far too late. They are scoped to an account, not to the project or the session that caused the spike, and they cannot stop anything.

Token counters

Show you the meter

Useful for curiosity, limited for control. Watching a number climb does not stop the next request from being sent.

Observability platforms

Analyse after the fact

Rich analysis, but it arrives after the spend, and it usually means sending prompts and responses to a third party to store.

Guardrail

Stops the request

Runs on your machine, in the path, with your own credentials. Policies are evaluated before each call, so an over-budget request is never sent and never billed.

Pricing

Free on your machine, forever.

The firewall, the policies and the local dashboard cost nothing and do not expire. You pay when you want the hosted dashboard, longer history, or team controls.

$0

Free

Local firewall, budget policies, local dashboard, 7 days of history. No account required.

$15

Pro, per month

Everything in Free, plus the hosted dashboard, 90 days of history and sync across machines. $144 yearly.

$99

Team, per month

A flat bundle for up to 10 seats, with an admin console, per-member spend and shared project caps.

FAQ

The questions everyone asks first.

Do you store my prompts or code?
No. Guardrail records metadata only— model, token counts, cost, timings, status. Prompt text, responses, code and credentials never touch its logs or database; a privacy layer and disk-scan tests enforce it.
Do I have to give Guardrail an API key?
No. It relays your existing Claude Code / Codex login unchanged — keyless passthrough. The only optional key powers the advisory Suggestions feature; it is stored locally and sent only to the model that grades your turns.
Can it actually block spend, or just watch it?
It blocks. Budget policies are checked before the request leaves your machine. An over-budget or denied-model call is stopped, no upstream call is made, and your tool gets a clean provider-shaped error instead of a crash.
Which tools does it cover?
Claude Code and Codex today, both fully metered and enforced, plus agent frameworks that speak the OpenAI Chat Completions API (LangChain, CrewAI, the OpenAI Agents SDK) in preview. Cursor is in progress via Cursor Hooks. See supported tools.
Is the cost my real provider bill?
It is an estimatefrom a configurable price table, shown with confidence labels — a close guide, not your invoice. Reconcile against provider billing for the exact number.
What happens if I stop Guardrail?
Start it again with guardrail start and your tools carry on. While the daemon is stopped it is still in the request path, so bring it back up before running an agent; a fail-open relay that passes traffic through untracked is on the roadmap.
Where does my data live, and what about Windows?
In a local SQLite file under ~/.guardrail, bound to localhost — nothing is uploaded unless you turn on cloud sync. It runs on macOS, Linux, and Windows through WSL2.
Is there a free plan?
Yes, and it does not expire. Free gives you the local proxy, budget policies, blocking, and the local dashboard with 7 days of history, with no account required. Paid plans add the hosted dashboard, 90 day history, and sync across machines.
Is pricing per developer or per workspace?
Pro is per person ($15 per month, or $144 per year). Team is a flat bundle, not a per-seat rate: one price covers up to 10 or 25 seats, so adding a teammate inside your bundle costs nothing extra.

Put a runtime budget around your AI agents.