NPU LabsNPU LABS
All products

RapportIQ · by NPU Labs

Relationship intelligence, built for the way you actually communicate.

RapportIQ turns the messages, emails, and chats your team is already exchanging into structured signal. Trust scores, risk detection, lifecycle phase, next-best action. Plug it into any AI agent via the Model Context Protocol.

The platform

Live
  • Channels

    6 sources

  • Extract

    Tone, intent, topics

  • Memory

    Long-term context

  • Lens

    Trust, risk, phase

  • Graph

    People, orgs, topics

  • Pulse

    Health, trends

  • Coach

    Next-best action

What it does

Every conversation, scored. Every relationship, watched.

Most CRMs ask you to log what happened. RapportIQ already knows. Communication arrives from the channels you already use, gets parsed for entities, topics, and intent, lands in a long-term memory keyed to people and organizations, and feeds an analysis layer that scores tone, trust, behavior, and risk. The result is a live read on every relationship and a coach that tells you what to do next.

Inside RapportIQ

Seven modules, one platform.

01 /

Channels

Pulls in every conversation from the apps you already use.

02 /

Extract

Reads each message for what it really says: entities, topics, intent.

03 /

Memory

Remembers everyone, every promise, every shift in tone.

04 /

Lens

Scores every interaction for trust, risk, and emotional weight.

05 /

Graph

Maps how your people, deals, and topics are actually connected.

06 /

Pulse

Surfaces the relationships drifting away before they break.

07 /

Coach

Tells you who to message and what to say, right now.

Where it listens

Meet people on the channels they already use.

RapportIQ ingests from the messaging, email, and chat sources that carry real conversations. No new tool for users to adopt.

Personal

  • WhatsApp
  • iMessage

Workplace

  • Slack
  • Microsoft Teams

Email

  • Gmail

Community

  • Discord

Built for AI agents

Plug in. Don't integrate.

RapportIQ speaks the Model Context Protocol natively. Connect Claude, GPT, or your own agent. They all use the same six tools to resolve people, log conversations, and pull relationship intelligence. No bespoke adapters. No drift between your AI and your data.

  • Streamable HTTP transport. Standard MCP SDK.
  • Channel agents stay thin. Domain rules live in one place.
  • Idempotent ingestion. Replay-safe by design.

Tool surface

v1.x SDK
  • create_organizationRegister a canonical account record.
  • create_personAdd a contact with channels and identity.
  • find_person_by_contactResolve an email, handle, or number to a person.
  • find_relationship_by_contactsMap two contacts to a relationship.
  • create_communicationIngest a message under the canonical pipeline.
  • list_supported_channelsDiscover what's accepted.

Six tools. Streamable HTTP. Official MCP SDK. Drop your agent in, start operating.

Grounded in research

Built on the work of others.

We are a lab. What we build is grounded in published research.

  1. [01] EACL · 2024

    RAGAS: Automated Evaluation of Retrieval Augmented Generation

    Es et al. (2024)

    Backs: Summary and suggestion gates. Faithfulness and AnswerRelevancy are RAGAS metrics, shipped via DeepEval.

  2. [02] EMNLP · 2023

    G-Eval: NLG Evaluation Using GPT-4 with Better Human Alignment

    Liu et al. (2023)

    Backs: Lens scoring. G-Eval is the general-purpose LLM-judge metric the evaluation worker runs.

  3. [03] NeurIPS Datasets and Benchmarks · 2023

    Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena

    Zheng et al. (2023)

    Backs: LLM-as-judge formulation. How Extract, Lens, and Coach scores are produced.

  4. [04] arXiv · 2024

    LLM Evaluators Recognize and Favor Their Own Generations

    Panickssery, Bowman, Feng (2024)

    Backs: Self-preference mitigation. The startup check that refuses judge model equals chat model.

  5. [05] arXiv · 2023

    Towards Understanding Sycophancy in Language Models

    Sharma et al. (2023)

    Backs: Bias mitigation. Neutral framing in custom-metric criteria avoids sycophantic judge behavior.

  6. [06] ICLR · 2020

    BERTScore: Evaluating Text Generation with BERT

    Zhang et al. (2020)

    Backs: Offline regression batches. Reference-based scoring against an authored calibration set.

  7. [07] IEEE TPAMI · 2018

    Efficient and Robust Approximate Nearest Neighbor Search Using Hierarchical Navigable Small World Graphs

    Malkov and Yashunin (2018)

    Backs: Memory module. pgvector and HNSW is how relationship memory is searched at scale.

Citations include preprints alongside peer-reviewed work. We interpret current LLM-agent research alongside established cognitive science, databases, and information retrieval.

More from NPU Labs.

RapportIQ is one of the things we build. See what else.