Back to homeFor developers and AI teams

Do your AI agents make decisions? Give them a method.

Static-rule guardrails block or let through, without understanding context. DAMM is the decision guardrail for your AI agents: it evaluates every single decision in context — the same action can be safe or catastrophic depending on who requests it, when, and why.

View API Docs Contact us for API access

What DAMM does that rules can't

Any team can write if-then rules. But rules are blind to context — the same action can be safe or catastrophic depending on who, when, and why. DAMM adds a decision intelligence layer: real AI agent guardrails that reason about context, not just static thresholds.

Context, not thresholds

A static rule says "block above €1,000." DAMM evaluates who's asking, when, with what history — and the same amount can get opposite verdicts.

Alternative actions

Rules just block. DAMM proposes the Minimum Move: the smallest action that produces a useful result without the risk.

Dynamic asymmetry

Not "how much you lose" but "how much you lose vs how much you gain, for this specific decision." Every verdict has a calculated risk/benefit ratio.

Readable audit

Not a technical log but a report that explains why. EU AI Act Art. 50 compliant without extra work.

See how DAMM helps with the EU AI Act

How it works

Three steps to integrate the DAMM method into your AI agents.

Agent sends

Your AI agent sends the decision context, available options and constraints through a simple API call.

DAMM analyzes

The DAMM engine applies the 4 pillars — Delimitation, Asymmetry, Margin, Minimum Move — and generates a structured verdict.

Verdict + Report

The agent receives the verdict, risk level, confidence score and a complete report for the audit trail.

Why static rules aren't enough

Two €400 refunds. Same amount, same agent. Static rules: same result. DAMM: opposite verdicts.

STATIC RULES

€400 refund — rule: "approve under €500"

1Customer A requests €400 refund → under threshold → approved
2Customer B requests €400 refund → under threshold → approved
3Customer A was a loyal 3-year customer, first return ever
4Customer B was a serial returner with 11 returns in 4 months
5The rule can't tell the difference: same amount = same result, always

Avoidable €400 loss + dangerous precedent

WITH DAMM

Same €400 — context changes everything

1Customer A (loyal, first return): Favorable asymmetry — refunding strengthens the relationship. Health Score 82 → APPROVE
2Customer B (serial returner): Unfavorable asymmetry — refunding reinforces the pattern. Health Score 31 → BLOCK
3Minimum Move for B: propose product replacement instead of cash refund
4Full report for both: who, when, why, with which alternative
5Manager sees two opposite verdicts for the same amount — and understands why

Same amount, different contexts, right verdicts

Simple Integration

A single REST endpoint. Send context, receive a structured verdict.

analyze-decision.ts

// Integration example (Coming Soon)
const response = await fetch('https://api.dammframework.com/v1/analyze', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_TOKEN',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    context: "Expansion into the German market",
    options: ["Open branch office", "Local partnership", "E-commerce only"],
    constraints: { budget: 500000, timeline: "6 months" },
    urgency: "medium"
  })
});

// Structured DAMM response
const { verdict, healthScore, riskLevel, report } = await response.json();

200 OKStructured response

// Response from DAMM API
{
  "verdict": "PROCEED WITH CAUTION",
  "healthScore": 58,
  "riskLevel": "medium",
  "pillars": {
    "delimitation": { "score": 7, "note": "Budget defined, tight timeline" },
    "asymmetry":    { "score": 5, "note": "Max loss 500K vs potential market 2M" },
    "margin":       { "score": 6, "note": "Reserves sufficient for 8 months" },
    "minimumMove":  { "score": 5, "note": "Can test with e-commerce only" }
  },
  "recommendation": "Start with e-commerce for 3 months before investing in branch",
  "auditTrail": "damm-report-2026-07-13-a8f3.pdf"
}

Read the full API documentation

Why an AI agent needs DAMM

Contextual Evaluation

A €400 refund is safe for a loyal customer and catastrophic for a serial returner. DAMM evaluates context, not just amounts — something no set of static rules can do.

Alternative Actions

Rules say "block" or "pass." DAMM suggests what to do instead: a smaller order, a test before rollout, a partial proposal. It doesn't stop the agent — it makes it smarter.

Structured Audit Trail

Not a technical log ("action blocked by rule #47") but a readable report: why this decision was risky, on which dimensions, with which alternative proposed. Native EU AI Act compliance.

DAMM vs static rules

Any team can write if-then rules. Here's what they can't do.

Static rules

DAMM

Same action, different contexts

Same result always

Different verdicts based on context

Blocked action

Stop. No alternative.

Proposes alternative Minimum Move

Risk/benefit ratio

Not calculated

Asymmetry calculated per decision

Audit trail

"Blocked by rule #47"

Readable report: why, on which dimensions, with which alternative

Adaptation over time

Manual: rewrite the rules

Automatic: context is dynamic by nature

EU AI Act compliance

Must build reporting separately

Native Art. 50 report in every response

A question of level

Three levels of guardrail. One governs decisions.

Content filters read the words. DAMM weighs the consequences.

Content guardrails

Filter the text: prompt injection, toxicity, sensitive data, jailbreaks. They look at what is written. Today a commodity: mature, often open-source libraries.

Conversational guardrails

Keep the dialogue on-topic and within an expected flow. They look at how the conversation goes.

Decision guardrails — DAMM

They weigh the action in context: safe for whom, and now? Not the text, but the consequences. The level the others don't reach.

The first two filter input and output. The third weighs the decision.

Build it, or adopt it

An LLM and a good prompt are enough for a prototype. The distance between a prototype and a system you answer for is measured over time.

A method, not a prompt

A framework structured in four pillars, not a string to rewrite at every exception.

Consistency

A tuned, versioned engine applies the same judgment every time. A prompt doesn't.

Compliance included

Audit trail and EU AI Act (Art. 50) transparency don't improvise themselves. Here they come as standard.

Maintenance

Every new model puts your prompt back in question. On this side of the line, it stops being your problem.

A rule takes an afternoon. Answering for a decision takes a method.

DAMM in action

Real scenarios where DAMM prevents high-cost agent errors. Each example shows the 4-pillar analysis and Health Score.

Finance

Algorithmic trading agent

Health Score

Scenario

Agent detects an arbitrage opportunity and wants to execute a €200,000 order. Market is volatile, spread is anomalous.

DAMM Response

Delimitation blocks orders above €50K without approval. Asymmetry detects expected gain (€3,200) is disproportionate to potential loss (€200,000). Minimum Move: test order of €10,000.

BLOCK — immediate escalationAuto-generated by DAMM

HR & Recruiting

CV screening agent

Health Score

Scenario

Agent must filter 500 applications for a senior position. Automatic filter is about to discard a candidate with unconventional but highly relevant experience.

DAMM Response

Asymmetry reveals: discarding an excellent candidate (opportunity cost: €150K/year) is far worse than passing a mediocre one (cost: 30 min interview). Minimum Move: create a "maybe" pool instead of discarding.

PROCEED WITH CAUTION — broaden criteriaAuto-generated by DAMM

Legal & Compliance

Contract analysis agent

Health Score

Scenario

Agent analyzes a liability limitation clause in a €2M contract. Clause appears standard but contains an unusual jurisdiction.

DAMM Response

Delimitation: agent is not authorized to validate clauses with jurisdictions not in the approved database. Margin: contract deadline in 72 hours — time for human review. Minimum Move: flag clause, propose standard alternative.

BLOCK — request legal reviewAuto-generated by DAMM