Exelab builds it for the customer
Exelab designs and delivers the agents; the customer uses them from day one.
See services → PRODUCT · POLYANT
Build AI agents that scale the enterprise.
Polyant is the platform designed to build AI agents that work alongside the enterprise team. Built for organisations that want to accelerate their processes with AI without giving up control, security, and data governance.
A platform to build tailored AI agents.
With Polyant you can build agents that work inside the company's processes, on concrete and defined tasks.
Converse naturally
Agents that handle complex conversations across existing enterprise channels, with context memory.
Check content
Automatic review before a message reaches the recipient.
Summarise and prepare briefs
From long documents to operational summaries in seconds.
Route incoming requests
Automatic classification and routing to the right team or system.
Extract information from documents
Structured reading of contracts, reports and forms, with output ready for enterprise systems.
03 · FEATURES
Sixfeaturesthatmakethedifference.
01 Runs where it's needed On-premise, customer cloud, or Exelab infrastructure.
02 Multiple agents that work together Squads of specialised agents coordinating on a shared task.
03 Converse across many channels Telegram, Slack, WhatsApp, OpenAI-compatible HTTP native in the runtime.
04 Persistent memory Every agent remembers the context of previous conversations.
05 LLM model of choice Anthropic, OpenAI, Mistral, or open source via AWS Bedrock.
06 Building blocks Tools, triggers, guards, routers: composable components for every workflow.
01 · ON-PREMISE
Your servers
The binary runs in your data centers. Data never leaves.
Polyant.agent 02 · CUSTOMER CLOUD
Customer cloud
Customer's dedicated account. Same agent, customer's perimeter.
Polyant.agent 03 · EXELAB INFRA
Exelab infra
Managed by the Exelab team. SLA and access governed by contract.
Polyant.agent TASK · IN FLOW
Process refund
#2849 · €420 · urgent
TR
Triage
classify · route
RG
Search
RAG · policies
DR
Draft
compose reply
QA
Approve
human-in-loop
01
Telegram
@user · bot · @user
02
Slack
@user · @polyant
03 WhatsApp
media · text · text
polyant.runtime channels.* → agent
HTTP · OpenAI-compatible POST /v1/chat/completions
polyant.support Session #14 · resumed CONTEXT LOADED
RECALL · 4 ENTRIES
- MAR 04 · 14:22 prefers email “please don't call me”
- MAR 18 · 09:51 order · SKU-2849 · €420 shipped to Milan
- APR 02 · 11:07 tone · concise, formal no emoji
- TODAY · 16:40 “Can I cancel the order?” agent remembers — answers in context
Claude anthropic / sonnet-4.5
ENGAGED CARTRIDGE BENCH
02 · READY
GPT-4o
openai
03 · READY
Mistral L
mistral.ai
04 · READY
Llama 3
aws bedrock · open
- Trigger /refund
- Guard auth · pii
- Tool erp.lookup
- Router auto / human
- Tool stripe.refund
04 THREE WAYS 01 Customer technical team 02 Customer business team 03 Customer developers
Three ways to work with Polyant, depending on who builds.
The internal team, via admin panel
Web interface to define roles, tools, memory, and channels without writing code.
Open quickstart →The internal team, via TypeScript
End-to-end TypeScript SDK for those who prefer full control of the code.
See docs →Getting started with Polyant
A few example agents
The catalog of ready-made agents for every department and industry.
Open the documentation
Quickstart, architecture, governance: everything you need to start building.
Talk to the team
A 30-minute conversation to understand whether Polyant solves a real customer case.
Frequently asked questions about Polyant
Polyant is a platform to build squads of AI agents inside the enterprise, not a single pre-packaged bot. Role, tools, memory, channels and policies are configured per agent; several coordinated agents cover an end-to-end business process.
Open the agent catalog for concrete examples by department and industry.
An agent has a defined role (e.g. AML screening, sales briefing, complaint classification), one or more tools to act on enterprise systems, persistent memory of the context, and built-in controls that step in before, during, and after every decision. Every step lands in the audit registry, queryable via standard SQL.
Anthropic Claude, OpenAI, Mistral, and open source models via AWS Bedrock. The model is hot-swappable: it changes without redesigning the agent, and each agent can use a different one. Model choice and execution region are part of the instance configuration.
Yes. Three deploy modes: on-premise in the customer datacentres, the customer's dedicated cloud account, or managed by Exelab in an EU region. The same agent runs in the three targets without changes. Perimeter is the customer's call and follows their data-residency rules.
Three modes, chosen by the customer: Exelab builds them in delivery on the real case; the internal team configures them via admin panel with no code; the customer's developers implement them via the end-to-end TypeScript SDK. The three paths produce agents compatible with the same runtime and the same audit registry.
Polyant has no public price list. Pricing is defined during qualification on the customer's real case: number of agents, execution volumes, managed profile, optional professional services. A 30-minute conversation is enough to get a concrete range.
From a 30-minute conversation to a team in production.
A 30-minute conversation to see whether Polyant solves a concrete customer case, and how long it takes to ship the first agent in production.