How it works
Defenses
Attack Patterns MCP Attack Atlas What we catch Hardening manual OWASP LLM Top 10 MITRE ATLAS
Learn
Encyclopedia (new) Agent Security 101 Blog Reports CVP runs Thesis
Resources
Docs Warroom vs Lakera vs Promptfoo Team
Theme
Pricing

Community is free. Boundary Review is the paid review.

Three offers · one engine · MIT stays MIT
Free
Community, MIT
$4,000
Boundary Review, per workflow
7
Business days
3
Routes at most

Community is free. Boundary Review is a paid review. Pro is for design partners.

The offers

Three Offers

community

Community

The scanner with its 1584 patterns, the CLI, the MCP server and the firewall hook. pip install sunglasses. free · MIT

review

Sunglasses Boundary Review

One agent workflow, reviewed at its trust boundaries. $4,000 per workflow, up to three routes, 7 business days. Founding offer for the first 5 buyers: $2,500. paid review

pro

Pro

A team evidence service for teams that need continuous receipts. No price and no product screens yet. Tell us what you need: [email protected] (subject: Pro). design partners wanted

sunglasses://community, what stays free
The promise

The maintained MIT Community engine, every right already released, honest fixes and independent local verification will not be degraded to create a paid tier.

Sunglasses Boundary Review

What Boundary Review Is

scope

Scope

One named agent workflow, up to three agreed routes, synthetic probes only. A larger scope is quoted on the scope call. The scope call is free.

time

Time

7 business days, counted from your written confirmation that working access is in place. A delay on your side pauses the clock. Includes one re-run of the same routes within 30 days of delivery.

you_get_routes

You get: the routes

Per route: MEDIATED, DIRECT or UNVERIFIABLE, with the basis. Per check: observed, failed or not run.

guard_down

The guard-down line

One per route: the checker is given a fault on purpose (for example a missing interpreter or a stall) and the line records what came back. Where the host's response was not exercised, it reads NOT RUN with the reason. Reported, not billed: it never changes the price, a refund or which routes count as exercised.

you_get_evidence

You get: the evidence

Receipts and verifier output you can check yourself, a fix list ordered by exposure and a one-page Method & Handling note. A completion letter is included: not a certification, not a compliance attestation.

you_do_not_get

You do not get

No grade and no pentest: not a certification, not an audit opinion, not a compliance claim.

payment

Price and payment

$4,000 per workflow, paid 50% to schedule (the deposit) and 50% on delivery (the balance). The price includes card fees, with no surcharge. The first 5 buyers pay $2,500, a discount off $4,000, for a quotable anonymized case study.

exercised

What counts as tested

A route counts as exercised when a benign control reaches the named tool from the running agent and returns a matched result, and a hostile control is attempted by the running agent at the agreed boundary and is either denied there with a matched receipt or reaches the server, which is reported as a finding.

not run

When the agent will not try

An agent that refuses to attempt the hostile control is NOT RUN. Hand-fed calls do not count. A route that misses either control is NOT RUN for billing, whatever label the findings table gives it.

refund

If no route is exercised

If no route is exercised, the deposit is refunded in full within 5 business days and the balance is waived. After a zero-route refund the engagement is closed and the re-run does not apply; a later review is a new engagement at the published price.

not run

If some routes are not run

Each agreed route reported NOT RUN removes an equal share of the delivery payment: the balance divided by the number of agreed routes. The amount payable is rounded down to the cent once, on the final total.

re-run

What the re-run covers

The re-run covers the same routes on a version or configuration you name, and delivers a revised findings table and a revised letter. The re-run must start, with access confirmed in writing, within 30 days of delivery.

re-run

A route not run at delivery

A route that was NOT RUN at delivery may be exercised in the re-run once access is in place; its share of the delivery payment is then due.

Buying a Boundary Review

How A Larger Scope Is Bought

What counts as a larger scope?

The standard review is one named workflow and up to three agreed routes. More routes, a second workflow or a longer access window is a larger scope. It is quoted on the scope call, and there is no public price for it. A change after the scope sheet is signed needs a written change order with the new price and date before work starts.

What happens on the scope call?

The call is free. The seven-business-day clock starts when you confirm in writing that working access is in place, and a delay on your side pauses it. We go through an intake: the workflow, the runtime and version, up to three routes, the sandbox or staging environment we may test in, who grants access and who to call if we must stop. The answers go on a scope sheet. Nothing is tested until the scope sheet is signed and you have given written permission for the synthetic probes.

How is payment handled?

50% to schedule and 50% on delivery. The price includes card fees. No surcharge.

What do we hold at the end?

A completion letter is included in the price: not a certification, not a compliance attestation. The report behind it states what was observed on the named routes at a stated time. It does not say every path was checked.

Boundary Review

Ask for a scope call

Write to [email protected] with the subject Boundary Review.