Promethist Relational Intelligence Engine

Build agents that know what to do next.

Built to improve outcomes, not just interactions.

The engine behind Empower and Engage, and the technology to build, run and manage your own Relational Agents. It keeps track of goals and agreements, chooses the next useful move, and speaks through voice, text or a Digital Human.

Engine traceEmpower · Rehearsal · Turn 6
Relational state
GoalHandle the price questionAgreed
Last sessionAnswered cost with featuresFact
ConfidenceLow on pricingInferred
Relational Judgment · next move
AnswerAskExplainChallengeRehearseWait
Coach"Let's run it again. Diane will ask about cost first. This time, lead with the number."
EvaluationAfter the run: did they lead with the price?

Illustrative trace. Field names simplified.

SecurityISO 27001 certified for information security
Amazon Alexa PrizeFour-time finalist, gold in 2021
What the engine does

Four parts. One loop. Every turn.

Recalling a conversation is easy. Choosing what to do with it is the hard part. The engine runs the same loop on every turn, in every product built on it.

  1. 01
    Relational stateWhat the agent knows about the person, the conversation and the goals, and what has changed. It tells facts and agreements from guesses, and doesn't treat them as equally reliable.
  2. 02
    Relational JudgmentChooses the next useful move: answer, ask, explain, challenge, rehearse, wait or bring in a person. It draws on the state, approved knowledge, your methods and your rules.
  3. 03
    AdaptationWhen the understanding changes, the next conversation changes with it. What is remembered only counts when it changes what the agent does next.
  4. 04
    EvaluationChecks whether the move did its job, so your team can see what works and change what doesn't.

Memory is the means. Judgment is the point.

Judgment, step by step

Same question. Three right answers.

A customer asks: "Should I move to the bigger plan?" Pick what the agent knows, and see what it decides.

Relational state
UsageOver the limit in each of the last three monthsFact
GoalStop paying overage feesAgreed
StylePrefers short answersInferred
Next move
AnswerAskExplainChallengeWaitBring in a person
Agent"Yes. You went over your limit three months in a row, so the bigger plan would have cost you less."
Why

The facts are clear and the risk is low. Another question would only slow them down.

Relational state
UsageNew customer, no history yetUnknown
MentionedA trip abroad next monthSaid
GoalNot statedUnknown
Next move
AnswerAskExplainChallengeWaitBring in a person
Agent"Happy to check. Is this about more data at home, or about your trip next month?"
Why

The right answer depends on something it doesn't know yet. One good question beats a wrong recommendation.

Relational state
AccountBusiness contract, 40 linesFact
RuleContract changes need the account managerRule
MoodFrustrated after two callsInferred
Next move
AnswerAskExplainChallengeWaitBring in a person
Agent"That change needs your account manager, Jana. I've sent her what we covered, and she'll call you today."
Why

Some decisions belong to a person. Knowing when to hand over, with the context, is part of the judgment.

Asking more is not always better. A clear, low-risk question deserves a straight answer.

What teams build on it

Bring your product. Add the judgment.

Relational Agents are AI counterparts designed to help people make decisions, develop abilities, reach goals and receive useful guidance in context. Empower and Engage are two of them. Teams with their own product can build on the same engine.

A Digital Human guiding a customer
Customer guidance agentsOnboarding, service and guided decisions, with a clean handover to a person when it matters.
Marco, a role-play client
Training simulationsRole-play partners and coaches for your own programs, built on the engine that runs Empower.
A Digital Human on a video call inside an app
Relational features in your appConversations inside the product you already run, that keep the context that matters, with the user's permission, and know when to act on it.
Customer story · TelecomHow a telecom operator uses the engine.Relational Agents built on our engine for business customer care that leads into sales.
Read the telecom story →
How it fits together

One engine, between your knowledge and your customers.

Your channelsApp · website · kiosk · phone
Voice and Digital HumanOptional, when a face or a voice helps
Relational AgentsEmpower, Engage or your own
Promethist Relational Intelligence Engine
Relational stateRelational JudgmentAdaptationEvaluation
Approved knowledge, methods and rulesSet and approved by your team
Operating controlsLimits, handover, monitoring
StudioConfigure, test and release agents: their goals, knowledge, methods and limits.
Channels and APIsSDKs and APIs for your apps, website and kiosks.
EvaluationSee how conversations went and whether they reached their goal.
Research and team

Built by the team behind four Amazon Alexa Prize finals.

We reached the final four times and won gold in 2021. The same team now builds agents for banking and telecom, from offices in New York and Prague.

4Alexa Prize finals
2021Alexa Prize gold
ISO 27001 certifiedAgents answer from approved knowledge, inside limits you set, and hand over to a person when they should.Compliance details →
Use cases

What teams build on the engine.

Relational Intelligence APIBuild agents that keep track of goals and agreements and choose the next useful move, on the engine behind Empower and Engage.Explore →Custom agentsYour agent, your knowledge, your rules, on the Promethist Relational Intelligence Engine.Explore →Enterprise deploymentHow the Promethist engine is deployed, secured and governed for regulated enterprises, and what to ask us for your security review.Explore →Data and knowledge integrationConnect agents to your approved knowledge and systems so every conversation draws on what your company knows, within the rules you set.Explore →One agent across channelsRun one Relational Agent on kiosks, websites, apps and phones, with the conversation and its context carried from one channel to the next.Explore →Research and labsUniversities, labs and enterprise innovation teams use the Promethist engine to study how people talk, learn and decide with AI agents.Explore →
Questions engineers ask

About the engine.

What is the Promethist Relational Intelligence Engine?

It is the engine behind Empower, Engage and every Relational Agent built on our platform. It keeps track of what matters about a person and a conversation, chooses the next useful move, adapts when its understanding changes and checks whether the move worked.

Do we need Empower or Engage to use it?

No. Empower and Engage run on it, and teams with their own product can build on it directly.

Which models do you use?

We use models from established providers and choose the right one for each part of the conversation. Your security team gets the current list on request.

Where can it be hosted, and who controls the data?

In our cloud, in the EU or the US. If you need another region, talk to us. Your company decides who can set up your agents and who can talk to them, with roles and single sign-on.

Which certifications do you hold?

PromethistAI is ISO 27001 certified. Security documents are available on request.

How do we integrate?

Through SDKs and APIs for your apps, website and kiosks.

We think the companies that grow most will be the ones that relate best.

Talk to our engineers
Bring one use case and your constraints. We'll show you how the engine would handle it.