AgentShield
How it works
All use cases Snowflake Cortex security Gemini Enterprise security Bedrock AgentCore security Azure AI Foundry security OpenRouter security AI gateway LLM security RAG security MCP server security Coding agents Customer service agents AI agent governance AI compliance Government Financial services Enterprise agents AI firewall Agentic AI security AI agent monitoring AI-SPM Agent hardening Tools compared Agent access control Agent identity AI data leak prevention Human in the loop Agent sandboxing AI penetration testing AI agent guardrails Healthcare Legal Insurance Multi-agent security Zero trust for agents Threat modeling AI runtime security MCP gateway LangChain security Claude agent security OpenAI agent security OpenClaw security Hugging Face security Cursor AI security n8n security Copilot Studio security LiteLLM security Gemini CLI security GitHub Copilot security Microsoft 365 Copilot security Kiro security Codex security Microsoft Agent 365 security OpenAI Agent Builder security Microsoft Agent Framework security Google ADK security Flowise security NeMo Agent Toolkit security Zendesk AI agents security ChatGPT workspace agents security Amazon Quick Suite security Oracle AI Agent Studio security Pydantic AI security Mastra AI security Dify AI security Notion Custom Agents security Zapier MCP security Gemini Enterprise Agent Platform security OpenAI Agents API security Claude Managed Agents security Copilot Cowork security ServiceNow AI agents security watsonx Orchestrate security SAP Joule security UiPath agent guardrails Security
Pricing Blog FAQ Contact Sign in

UiPath Agent Pricing and Agent Units Cost Compared with AI Units and Platform Units

AgentShield Security Team·Oct 15, 2026·7 min read

Try it live

Watch AgentShield block an attack in real time.

Pick a scenario and drive the inspection lane yourself. No signup needed.

Threat Console
Interactive demo · 0 blocked in this session

Run a request

Runs the live engine on your text. Nothing is stored, no account needed.

Inspection lane

INSPECTING
⌖ untrusted input

Policy trace

High-risk action held for approval

Audit trail

Compliance manager and security architect comparing options on a whiteboard in a meeting room

UiPath agent pricing is consumption based. Production agent runs consume Agent Units on the Flex plan or Platform Units on Unified Pricing, charged per LLM call on UiPath-hosted models: 1 Agent Unit or 0.2 Platform Units for a standard model, double that for premium models, and 0.8 of it for basic models. UiPath does not publish a dollar price per unit, and its 25 USD per month Basic plan can design agents but cannot run them in production.

The rates below were read on UiPath's agent licensing documentation and its public pricing page on October 15, 2026. UiPath changed several entitlements during 2026, so many third-party pricing pages still say the unit rates are not public. The rates are public. Only the dollar value of a unit is negotiated.

UiPath agent pricing at a glance

What you useFlex planUnified PricingWhat to watch
Standard models (GPT-5.x, Claude Sonnet, Gemini 3 Pro)1 Agent Unit per LLM call0.2 Platform Units per LLM callAn agent that makes eight calls per run pays eight times
Premium models (Claude Opus 4.7, GPT-5.5, preview)2 Agent Units per call0.4 Platform Units per callDouble cost, so reserve them for steps that need them
Basic models (Claude Haiku 4.5, GPT-5-mini, Gemini Flash)0.8 Agent Units per call0.16 Platform Units per callOften good enough for classification and routing steps
Your own model through the AI Trust Layer1 Agent Unit per run0.2 Platform Units per runFlat per run, but you also pay your model provider
Design and evaluation in Studio WebPer-user monthly usage pool, then top-ups, then unitsSame orderThe pool belongs to each named user and cannot be shared

One detail changes real bills. UiPath charges LLM calls in 64,000 input token increments, and input includes the system prompt, conversation history and tool definitions. A call with 100,000 input tokens counts as two calls. Agents with long instructions and many tools cross that line faster than teams expect.

Agent Units vs AI Units vs Platform Units

These names cause most of the confusion in UiPath agent quotes.

UnitPlanWhat it pays for
Agent UnitsFlexAgent runs in production, per LLM call or per run on your own model
AI UnitsFlexOther UiPath AI products. A separate consumable that does not cover agent runs
Platform UnitsUnified PricingOne shared pool that replaces the separate Flex consumables, including agents

The practical consequence on Flex: a tenant with plenty of AI Units can still fail to run an agent if no Agent Units are allocated. UiPath also says each tenant starts with 0 Agent Units and draws from the organization pool unless you allocate units to it, so decide early whether production tenants get a fixed allocation.

How to estimate UiPath Agent Units cost for one agent

Take an accounts payable agent that reads an invoice, checks the vendor, matches the purchase order, drafts an exception note and updates a record. Count six LLM calls per run on a standard model, and 20,000 invoices a month.

  • Flex: 6 calls times 1 Agent Unit times 20,000 runs is 120,000 Agent Units a month.
  • Unified Pricing: 6 times 0.2 times 20,000 is 24,000 Platform Units a month.
  • Same agent on your own model: 20,000 Agent Units or 4,000 Platform Units a month, plus your model provider's token bill.
  • Same agent with an LLM as Judge check on every call using a basic model: add 0.8 Agent Units per check, or 96,000 Agent Units a month on Flex.

Multiply by the unit price on your order form to get dollars. Then ask whether the job needs an agent at all. If the agent's real work is pulling header and line-item fields from invoices, a dedicated invoice data extraction tool priced per document is often cheaper than paying for six model calls per invoice, and you keep the agent for the exceptions.

What guardrails add to the UiPath bill

Most UiPath agent security costs nothing extra, and that is worth saying plainly before anyone sells you a separate product.

  • Managed guardrails are included on eligible plans. PII detection, prompt injection, user prompt attacks, harmful content and IP protection are available on Flex Enterprise Standard and Advanced, and on Unified Standard, Enterprise and the App Test plans.
  • Custom tool guardrails are free. Deterministic rules on tool inputs and outputs with Log, Filter, Block or Escalate add no unit cost, and they are the control most agents that write data need.
  • LLM as Judge is billed. It makes its own model call and consumes units at the same per-model rates, tracked separately from the agent. UiPath suggests a lower-tier judge model to keep the cost down.
  • Bring your own guardrails means a second subscription. Since October 1, 2026, administrators can connect Azure AI Language, Azure Content Safety or Noma in preview, and you pay that vendor directly.
  • Escalations need serverless robots. The guardrail escalation app must be deployed in a shared folder that has serverless robots available, so confirm your plan covers them.

One governance setting deserves a look at renewal. Since October 7, 2026, administrators can set a policy that lets users get AI and agentic usage top-ups automatically instead of requesting them. That removes a spending check, so decide who owns it.

The cost that does not appear in UiPath's SKU is a control for tools shared with agents outside UiPath. If the same payment, vendor or access API is called by UiPath agents and by Copilot, Claude Code or Codex, the rules have to live somewhere all of them pass through. That is what we sell. AgentShield is published at 49, 199, 749 and 1,498 USD per month on its four plans, half that when billed yearly, and every plan is bought by card. Our UiPath agent guardrails page shows exactly where that gateway sits and where UiPath's own rules are enough.

When the extra security spend is not worth it

DeploymentExtra security spendWhy
Agents that read, classify and summarizeNoneManaged PII and prompt injection checks are included on eligible plans
Agents that write through tools only UiPath agents useNoneCustom tool guardrails with Block and Escalate cover it at no unit cost
Agents where a policy needs judgment rather than a field ruleLLM as Judge unitsEach check is a billed model call, so keep it to the steps that need it
Payment, vendor or access APIs shared with agents on other platformsA per-call gatewayOne rule set and one approval record across every agent that calls them

Three of four rows need nothing from us. For how to choose which actions deserve a second person at all, see our guide to human approval for AI agents, and for how per-argument rules are written once for every agent, see tool permissions for AI agents. If you also run SAP, our SAP Joule pricing guide covers the AI Units side of that bill.

Questions buyers ask before signing

How much does a UiPath agent cost?

It depends on LLM calls per run, model tier and volume. On Flex a standard model costs 1 Agent Unit per call, and on Unified Pricing 0.2 Platform Units per call. UiPath does not publish a dollar price per unit, so the dollar figure comes from your quote. Count calls per run during a pilot, then multiply by expected monthly runs.

Are UiPath Agent Units the same as AI Units?

No. On the Flex plan they are separate consumables. Agent Units pay for agent runs in production, and AI Units pay for other UiPath AI products. A tenant with AI Units but no Agent Units cannot run agents in production. Unified Pricing replaces both with a single pool of Platform Units.

Do UiPath agent tests consume Agent Units?

Not at first. Design-time debugging and evaluations in Studio Web draw from each named user's monthly AI and agentic usage pool, then from administrator-granted top-ups, and only then from Agent Units or Platform Units. Production runs started from Orchestrator, an API, a webhook or a published conversation channel always consume units.

Can I run UiPath agents on the Basic plan?

You can design, debug and evaluate agents on Basic, which UiPath lists from 25 USD per month, but UiPath's licensing table does not include running agents as processes on Basic. Running agents in production needs Standard or Enterprise, which UiPath prices through sales. Check this before a pilot that is meant to go live.

Next step

Before you size units, list which agents will act and which will only read, and for the ones that act, which tools they write through. That list sets the unit forecast, the guardrails you configure in UiPath and whether any tool needs a control outside it. If your agents share APIs with other platforms, compare AgentShield plans and route those calls through one gateway.

See the firewall block an attack live.

Drive the Threat Console and watch a real prompt injection get stopped, then put AgentShield in front of your own agents.

Open the console