myuiaf

Live  Freshness Validator

Keep your AI agents from using outdated information.

Policies, prices and API documentation change, but your agent doesn't notice. Freshness Validator monitors the sources your agents rely on and tells them, in a single call, whether their information is still correct.

Example validation results
Context the agent holdsScore
Rate limit is 100 requests per minute
Source unchanged since last check
92
fresh
Support hours end at 5 pm
Validated 6 days ago, policy is 7 days
74
warning
Refund window is 30 days
Source page changed after last validation
58
stale

Why teams use it

Fewer wrong answers

Customers, colleagues and other systems get today's facts, not the version your agent loaded last month.

Fewer tokens

Instead of loading entire documents into every request just in case, the agent re-reads a source only when it has actually changed.

Less time fixing mistakes

No searching for incorrect answers after a policy change, and no debugging code written from outdated documentation.

Real-world examples

Support

Refund policy changes

WithoutThe window drops from 30 to 14 days. The bot keeps promising 30 days, so you either honor incorrect promises or handle complaints.
WithThe check marks the old answer as stale, so the bot re-reads the policy once and answers "14 days".
moneyreputation
Engineering

API documentation changes

WithoutAn endpoint is renamed. The coding assistant writes code from the old documentation, the build breaks, and someone loses an hour.
WithThe documentation page is monitored, so the change is flagged and the assistant uses the current version the first time.
developer timetokens
Compliance

A regulation is updated

WithoutA research agent keeps citing the previous rule in reports, and nobody notices until a reviewer does.
WithThe regulator's page is monitored, so anything checked before the change is flagged as stale.
riskreview time
Sales

Prices go up

WithoutA sales assistant sends a customer a written quote based on last quarter's prices.
WithThe price page change is detected, the quote is flagged before it goes out, and the assistant uses the current prices.
marginawkward emails

What would you save?

Enter your own numbers. The math runs in your browser.

Re-reading every time vs. checking first

Many agents load a whole document into every request "just to be safe". Enter your own numbers.

Re-read on every use-
Check first, re-read only on change-
-
AI tokens not spent per month-

Estimates from your inputs, per 30-day month. 1 page ≈ 800 tokens. Defaults are examples, not measurements. If you use prompt caching, enter your cached-token price.

What it does

Time-based scoring
Scores context by how long ago it was validated, against a freshness policy you choose: 1h, 24h, 7d, or a custom window.
Source monitoring
Fetches and hashes the sources you register on a schedule. When a page changes, context validated before the change is flagged.
Content matching
Compares your information with the current page text, five words at a time. Changes to the actual wording count; changes to layout, scripts and styling do not.
REST and MCP
Call it over standard REST from backend services, or over the Model Context Protocol from agents. Both use the same validation engine.
Pay per call
Prepaid credits are deducted for each check. If the wallet is empty, the call returns HTTP 402 with no charge and explains how to add credits.
Per-agent API keys
Each agent has its own key. Secrets are stored as SHA-256 hashes, shown once at creation, and can be revoked and replaced at any time. Repeated failed attempts are throttled by IP.

How it works

  1. Register sources. Tell the service which web pages your agents get their information from.
  2. Check information. Send the information your agent is using. You get a score, a stale flag and suggestions.
  3. Act on the result. If the information is stale, the agent re-reads the source and checks again.

See how scoring works

Try it

Register a source, then check information against it. Replace the example key with your own.

# Register a source for monitoring (costs 2 validation calls)
curl -X POST https://api.myuiaf.com/api/v1/source/register \
  -H "Authorization: Bearer fv1.acme-bot.k1.xxxxx" \
  -H "content-type: application/json" \
  -d '{"source_url":"https://example.com/api-docs","check_frequency":"6h"}'

# Validate agent context
curl -X POST https://api.myuiaf.com/api/v1/validate \
  -H "Authorization: Bearer fv1.acme-bot.k1.xxxxx" \
  -H "content-type: application/json" \
  -d '{"context":"The API rate limit is 100 req/min","source_url":"https://example.com/api-docs"}'
{
  "freshness_score": 72,
  "is_stale": false,
  "confidence": 0.85,
  "source_match": { "method": "overlap", "overlap": 0.95 },
  "flagged_issues": [],
  "suggestions": [],
  "source_status": "monitored",
  "credits_deducted": 0.03,
  "remaining_balance": 4.97
}

Full API reference and MCP setup

Questions

What does Freshness Validator do?

It scores how fresh an AI agent's context is, from 0 to 100, by combining how long ago it was validated, whether its source page has changed, and how closely the text still matches the source. Scores below 70 are flagged as stale.

Does it store the context I send?

No. Submitted context is scored in memory and not stored. Only the text of the source pages you register is kept, to detect changes.

Can I use it from an agent over MCP?

Yes. The same validation engine is available over REST at api.myuiaf.com and as an MCP server at https://api.myuiaf.com/mcp.

How much does it cost?

From $0.03 per validation call on the Early Testing tier down to $0.015 on Enterprise. Registering a source costs two calls. An agent with no account can pay $0.05 in USDC for a single call.

What happens when the wallet is empty?

The call returns HTTP 402 with no charge and no result, and tells the agent how to buy more credits or pay for the single call.

Works with AgentPay

The same API key and prepaid credits also work with AgentPay, our pay-per-call gateway to third-party APIs. One wallet covers both services.