Skip to main content

We found a way to make AI agents do exactly what you want

at roughly half the context cost

SIGN™ is the open standard that tells your agents what’s true, what’s allowed, and what to do about it.

Explore SIGN
View on GitHub

The Problem

Agents can access data. They can’t apply it.

Your rules already exist. They’re in a policy doc, a data model, someone’s head. Your agents can find them just fine.

They can’t apply them.

That’s not a retrieval problem. Retrieval works. Nobody built a place for a rule to live in a form a machine can actually use — so agents guess, and guessing at scale is where trust breaks.

agents-cannot-access-data

How It Works

SQL for enterprise knowledge

SQL won because it separated the data from how you use it. SIGN does the same for knowledge — facts, rules, exceptions, and who has the authority to change them, in one compact format an agent can reason over.

JSON stores data. Documents describe intent. Neither can be enforced. SIGN can.

EXAMPLES

@doc

BILL-CREDIT-014 | type: policy | status: active | audiences: billing-agent, finance-ops

## Determines when a billing dispute qualifies for automatic credit versus manual review.

@links

-> BILL-CREDIT-014-APPENDIX-A (escalation matrix)
<- CONTRACT-SLA-009 (source SLA terms)

@anchor

Master Services Agreement §7.2; Billing Policy Handbook p.14

@def

CREDIT-ELIGIBLE-DISPUTE
A billing dispute where the disputed amount and cause both fall within the automatic-credit threshold.

@props

max_auto_credit_amount: $500
dispute_window_days: 30
prior_credit_limit: 1 per trailing 12mo

@include

  1. Disputed within 30 days of invoice date

  2. Cause code is one of: duplicate-charge, service-outage, pricing-error

  3. Customer has no more than 1 prior auto-credit in trailing 12 months

@exclude

  1. Disputed amount exceeds $500

  2. Cause code is contract-termination or fraud-flagged

  3. Account has an open collections case

@rules

  1. If @include is fully satisfied and no @exclude condition is triggered, apply automatic credit.

  2. If disputed amount exceeds max_auto_credit_amount, route to manual review queue regardless of cause code.

  3. If prior_credit_limit is exceeded, escalate to finance-ops regardless of amount.

This is what the policy is, when it applies, and what can’t happen — in a form an agent can check itself against.

OPEN STANDARD

Open where adoption matters

A knowledge format only works if everyone can use it. If every company builds its own version, nothing interoperates and adoption stalls.

We’re releasing SIGN under the Appache 2.0 License for the same reason SQL and HTTP won: nobody had to ask permission to use them.Because relationships and logic are encoded directly, agents require less context to achieve the same outcome.

  • No lock-in — SIGN works natively, any agent, anywhere
  • Standards empower evaluation — agents get better as a result
  • Enterprise knowledge compounds
lower-cost

USE CASES

cut-ai-costs-with-sign

WHERE SIGN™ GOES TO WORK.

  • Code coverage
  • Agent instructions (agents.md, claude.md, etc.)
  • Corporate knowledge
  • Contract conformance
  • Business process

We run our own company on this. It’s the reason we built it.

See it, try it, or talk to us — whatever’s next for you.