SECURITY-FIRST AI ROUTER Coming soon

 

REQUEST PATH / 01Planned flow · Coming soon
YOUR APPLICATIONYour application
POST /chat
BOUNDARY POLICY GATEPolicies before transmission.
Planned
Check provider and model allowlists
Check defined sensitive-data patterns
Check the project budget
ILLUSTRATIVE DECISIONBLOCK secret_pattern_match

A defined pattern match would block transmission.

ALLOWED PROVIDERAllowed model provider

Illustrative planned flow, not a live request or product screen.

YOUR APP. YOUR POLICY.

Boundaries for your model requests.SCROLL TO EXPLORE

01 / BEFORE THE MODEL

Before the model,
a place to set limits.

Choosing a model is only part of the job. Boundary is being built to put transmission rules in the API layer, so your team can decide what goes where.

01

Choose where requests go.

Planned project policies will limit requests to providers and models you allow.

provider.allowlist
02

Define what stays out.

We plan to block requests that match defined secret or personal-data patterns before transmission.

secret / pii.patterns
03

Keep content out of logs.

We plan to leave raw-content logging off by default and record policy decisions and audit metadata.

raw_content.log = off

The initial scope is model requests sent through the gateway. Defined patterns will not detect every sensitive detail, and upstream provider retention policies remain separate.

02 / A LAYER, NOT A REBUILD

Keep your app.
Add a policy layer.

We’re building a model API router that brings security policies and project budgets into a shared connection layer.

See the planned scope

Boundary is an API router, not an agent runtime or a coding harness.

client.tsPlanned integration
// Illustrative integration; not a live API.
const client = new OpenAI({
  apiKey: process.env.BOUNDARY_API_KEY,
  baseURL: process.env.BOUNDARY_BASE_URL,
});

// Planned gateway policy and budget checks
await client.chat.completions.create({
  model: "your-allowed-model",
  messages,
});
Endpoints and SDK support will be confirmed before launch.

03 / INITIAL GATEWAY SCOPE

A focused start.
A clear scope.

All features below are planned.
They are not available in a live API today.

FeatureInitial scopeStatus
Provider and model allowlistsRoute requests to project-approved providers and modelsPlanned
Sensitive-data pattern checksCheck defined secret and personal-data patterns before transmissionPlanned
Raw-content logging off by defaultRecord decision reasons and audit metadata, not raw contentPlanned
Project budgetsManage model usage budgets at the project levelPlanned
THE BOUNDARY OF BOUNDARY

Tool execution controls, complete prompt-injection detection, and zero-retention guarantees across providers are outside the initial scope.

04 / EARLY ACCESS

An early thank-you
for early interest.

Join the launch list for updates and details on your first paid month’s discount.

Reserve my 50% discount
  • 50% off the security subscription for your first paid month after launch. Model usage fees are separate and are not discounted.
  • Joining the list is free. It is not a preorder, purchase, or paid subscription. No account or payment details are required.
  • Launch date and final pricing are not yet set. Features are planned and may change before launch.
BOUNDARY / EARLY ACCESS

50% off your first paid month of the security subscription

First paid month after launch · Model fees excluded

No payment details requiredLaunch date and pricing to be confirmed

05 / GOOD QUESTIONS

Good questions.
Clear answers.

Can I use Boundary today?

Not yet. We’re planning a model API router with provider and model allowlists, defined secret and personal-data pattern checks, raw-content logging off by default, project budgets, and policy decision metadata. These are planned features, not a live API. Launch date and pricing are not yet set.

What will the gateway protect?

Configured policies will apply to model requests sent through the gateway. They will not cover data that falls outside defined patterns, requests that bypass the gateway, or tool execution. We do not promise complete prompt-injection detection or zero data retention across upstream model providers.

What will you log without the raw content?

We plan to leave raw request and response logging off by default and record policy decisions and audit metadata instead. The exact fields will be published before launch. Each model provider’s own retention policies still apply and must be reviewed separately.

Is this a coding agent or an agent runtime?

Boundary is a model API router. It is designed to route model requests and apply transmission policies, not to plan tasks, edit files, host agents, or execute tools. Supported APIs and integration details will be confirmed before launch.

What is discounted, and do I pay anything now?

The launch offer is 50% off your first paid month of the security subscription. Model usage fees are separate and excluded. Joining the email list costs nothing and is not a preorder or paid subscription. Final pricing is not yet set.

How will you use my email?

With your consent, we use your email for launch and discount updates. Signup records are kept for up to 12 months from consent and deleted if you withdraw consent. Signup links your email to visit and campaign data for response measurement; measurement events are kept for up to 90 days. You may decline consent, but then we cannot send those updates. See the privacy notice for details and deletion requests.

Privacy notice

With your consent, we collect your email for launch and discount updates. Signup records are kept for up to 12 months from consent and deleted when you withdraw consent. You may decline consent; we cannot send updates without it.

We store a random visitor ID and page variant in your browser to measure responses. Measurement events contain no email, but signup records link your email to the visitor ID, page variant, and campaign data. These records can therefore be associated with your measurement history. Measurement events are kept for up to 90 days; signup records for up to 12 months. Only constrained campaign labels are stored. We currently use no third-party advertising pixels or automated email sending.

Data controller: OraXLabs, LLC · Withdraw consent or request deletion: hello@oraxlabs.io