Skip to content
Platform → AI agents

No single agent has the whole picture

Merivex does not run one general-purpose model over your data. The analysis is split across specialist agents, each with a short tool list and a narrow memory scope, and a verifier whose job is to disagree with the rest.

Coordinate

Ingest

Understand

Evaluate

Verify

Discover

Coach

Report

Govern

Coordinate

01 / 14

Supervisor / Router

Routes analysis tasks to the right specialist and coordinates multi-agent workflows.

Tools
3
Reads
organisation
Writes
organisation

The stages

Nine stages, one direction of travel.

Every agent belongs to a stage, and the stages run in order — coordinate, ingest, understand, evaluate, verify, discover, coach, report, govern. Nothing skips the verifier.

01Coordinate

Decides which specialist runs, and in what order. Holds no analytical opinion of its own.

  • Supervisor / Router
02Ingest

Gets the data into one shape and pulls the history a judgement needs.

  • Ingestion / Normalization Agent
  • Context & Knowledge Retrieval Agent
03Understand

Classifies what the conversation was about and how it moved, before it is scored.

  • Intent Agent
  • Sentiment / Emotion Timeline Agent
04Evaluate

Scores the interaction against your rubric and surfaces policy risk, every claim tied to a turn.

  • Evaluation Agent
  • Compliance & Risk Agent
05Verify

Re-reads the conclusions and rejects anything the cited evidence does not support.

  • Critic / Verifier Agent
06Discover

Rolls individual findings into patterns, and puts the human review queue in priority order.

  • Pattern Intelligence Agent
  • Review Triage Agent
07Coach

Turns the evidence into a specific, per-person plan.

  • Coaching Agent
08Report

Writes the executive narrative over live metrics. Reads only.

  • Analytics Agent
09Govern

Keeps memory clean and the connectors healthy.

  • Memory Curator
  • Data Quality / Integration Health Agent

Scoped authority

An agent cannot quietly become authoritative.

Each agent is restricted to an explicit set of tools and explicit memory read and write scopes. It operates inside the same tenant isolation and authorization model as a human user, with no bypass.

  • Short tool lists — most agents can call between three and nine tools, and nothing else.
  • Two agents (Analytics, Memory Curator) write nothing at all. They read and report.
  • A conclusion only becomes authoritative after the Critic / Verifier agent has failed to break it.
  • Merivex AI agents never write to a system you connect. Read-only, by design.

The authorization boundaries are described in full on the security page — what each agent may touch, and why a scoped toolset is a control rather than a limitation.

Read the security page →

Two kinds of agents

Your people, and the AI that studies their work.

Merivex uses the word deliberately in two places. They are never the same thing.

Intelligence → Agents

Your workforce

The people handling customer conversations. Merivex measures their work, tracks quality over time and builds individual coaching plans, organised by team, by interaction and by evidence.

Platform → AI Agents

Merivex's internal AI agents

Merivex's own analysis engine. Specialists for ingestion, evaluation, compliance, intent, sentiment, patterns, coaching, verification and memory, each with a scoped toolset and no authority beyond it.

See it against your own conversations

The agents above run over real interactions and cite the turn every conclusion came from. Tell us how your quality programme works today and we will show you what comes back.