What it does
8 features
UiPath Agents · Agent capability of an automation platform
Product and access
This profile only describes the agents. The docs distinguish native UiPath Agents, built in the platform, and Coded Agents, developed outside and deployed there. Robots (RPA), Maestro and Automation Suite are other platform parts the agents use.
Core features
8 features from provider sources · checked on 5 November 2026. Availability and limits are directly below.
Agent Builder in Studio Web
Autonomous or conversational agents are built in form or canvas view, from scratch or with Autopilot. You set model, temperature and maximum iterations.
Source 6Tools
An agent only acts through tools it is explicitly given access to, such as automations, API workflows, other agents or MCP servers.
Source 7Escalations
The agent hands decisions to people through Action apps in Action Center. For each outcome you set whether it continues or ends.
Source 8Guardrails
Built-in checks for personal data, prompt injection and harmful content can log, block or escalate. They depend on the licence plan.
Source 9Traces and evaluations
Traces show the steps, inputs, outputs, errors and token usage of a run. Evaluation sets test agents against defined inputs.
Source 10Agent memory (preview)
Episodic memory recalls similar earlier cases, escalation memory reapplies earlier human decisions.
Source 11
Show all 8 features
Coded Agents
Agents developed outside UiPath can be deployed on the platform and have their own licensing model.
Source 12Self-operation
According to the provider, Automation Suite brings the agent stack to your own infrastructure on AKS, EKS or OpenShift, optionally with models in your own data centre.
Source 13
Usable AI models
Open choice
In Agent Builder you choose from the models UiPath provides from OpenAI, Anthropic and Google. Through the AI Trust Layer, admins can also connect their own subscriptions or models, including via any OpenAI-compatible gateway; this route is for Enterprise customers on request. In Automation Suite, self-hosted models are possible, but only with part of the agent features.
Model families named
- OpenAI (GPT)
- Anthropic (Claude)
- Google (Gemini)
- Other models can be connected freely
- Model choice
- You choose
- Your own API key
- Yes
- Router or custom endpoint
- Yes
- Local models
- Yes
- Depends on
- Plan, Admin approval
The rating covers the generally available main access. Documented · Sources accessed on
UiPath: AI features and model routing · UiPath: Connecting open-source models · UiPath: Bring your own model · UiPath Automation Suite: LLM support
Does it fit my task?
2 practical tasks · 6 requirements and limits
A good fit if …
- you already run UiPath automations and want to add agents there
- human escalations, guardrails and traces are to be a fixed part of operation
- you have to run agents in your own Kubernetes environment
Less suitable if …
- you are looking for a personal assistant without platform build-up
- you want to estimate costs without sales contact and consumption calculation
- your workflow allows no mistakes and has no deterministic safeguards
Editorial assessment based on provider documentation. Not a product test.
Practical tasks
Editorial examples based on documented features. No product tests performed.
Pre-sorting tickets
- Your input
- Incoming requests and an agent with a knowledge index as context.
- Possible result
- The agent classifies the request and drafts a suggestion; unclear cases go to Action Center.
- What to check
- Create an evaluation set and check the traces of the first runs before the agent runs more widely.
Connecting an agent to an existing automation
- Your input
- An existing RPA process as a tool and a prompt with goal and limits.
- Possible result
- The agent decides when to call the process and reports the result.
- What to check
- Set guardrails for the tool and test with simulations before real systems are addressed.
Have an agent pre-sort incoming tickets and hand unclear cases to a person through Action Center.
Your possible prompt
Compare three providers for [task]. Use their official documentation. Show differences, sources, and open questions in a table.
Adjust the task, data and permissions.
Requirements and limits
Requirement
Running from Standard
On Unified Pricing, Basic plans can only design agents. They run as processes from Standard Trial, Standard and Enterprise.
Source 14Limitation
Not for every workflow
The docs name high-risk financial transactions, zero-error regulatory workflows and multi-step data processing without deterministic safeguards as poor fits.
Source 15Limitation
Differences by delivery
Dedicated offers no conversational agents and no agent memory and only part of the guardrails.
Source 16Limitation
Not in Public Sector
The product overview lists Agents as available for Automation Cloud, Dedicated and Automation Suite, not for Public Sector.
Source 2Not yet confirmed
Automation Suite on Linux
The product page names agent capabilities on Linux as planned for October 2026. Whether they have shipped was not confirmed.
Source 13Not yet confirmed
Retrying failed runs
The page on running names the failed status but describes no retry or recovery.
Source 17
Can I use it?
With requirements
Europe & regional availability
Region rules, model routing, guardrails page and Suite post were reviewed. No table of Agents per region was available. The data residency page only names the EU, model routing also the United Kingdom and Switzerland.
Reviewed scope: Automation Cloud and self-operated Automation Suite
Reviewed on:
With requirements
Data residency by plan
According to the page Community, Free and Basic store in the EU. Standard inherits the organisation's region, a new trial organisation can choose it. Enterprise chooses per tenant.
Region, reason and source
Region covered: EU
Documented reason: The page governs platform data and does not name Agents separately.
With requirements
Model routing for agents
Many models stay in the region for EU tenants. Some run outside, for example anthropic.claude-sonnet-5 via Bedrock in the US; global preview models have no fixed region. Model Regions can restrict this, and calls to unavailable models then fail.
Region, reason and source
Region covered: EU
Documented reason: Depends on the chosen model and the policy.
Official source · docs.uipath.com · Official source · docs.uipath.com
With requirements
Switzerland and United Kingdom
The routing table lists both as separate columns. Models without a regional endpoint route to the EU or the US, only a few stay in the country.
Region, reason and source
Region covered: Switzerland and United Kingdom
Documented reason: Depends on the model; the data residency page does not name either region.
With requirements
Prompt injection check in the US
The guardrail runs at Noma Security in the US, so data can be processed outside the tenant region.
Region, reason and source
Region covered: EU/EEA, United Kingdom and Switzerland
Documented reason: Service location named by the provider.
Available
Self-operation
With Automation Suite, configurations, data and audit trails stay in your Kubernetes environment according to the post. Models can run there as well.
Region, reason and source
Region covered: Location chosen by the operator
Documented reason: The post generally names sovereignty and in-country processing, but no individual countries.
Official source · www.uipath.com · Official source · www.uipath.com
Accounts & permissions
Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.
Configuration covered: Provider platform or your own Automation Suite
Sources reviewed:
Show all 3 points
Documented with conditions
Data access & team sharing
An agent only uses tools it is explicitly given access to. Through the AI Trust Layer policy, administrators control access per tenant, central guardrails and permitted models. The tools page does not describe which rights a tool has in the target system.
Evidence · Evidence
Memory and conversation continuity
3 of 6 criteria documented
Episodic and escalation memory in memory spaces that can be reviewed and edited. Preview, not on Dedicated; retention period not stated.
Agent memory as a preview, in shared memory spaces
Sources accessed:
Documentation by criterion
StoragePartly documented
Storage and scope
Memories sit in memory spaces holding items and match rules; they are assigned to folders. Customer-managed keys are set at creation. The page names no retention period. Available on Automation Cloud and Test Cloud, not on Dedicated.
SelectionDocumented
Selection and timing
You add episodic items manually by promoting feedback from a trace or an agent output; a check looks for duplicates first. Escalation memory saves automatically when a person resolves an escalation.
Edit & deleteDocumented
Inspect, edit and delete
Items can be reviewed, edited and deleted in the Memory spaces tab. The inputs of an episodic item cannot be changed; you delete and re-add it. Memory spaces can be exported.
Access boundariesPartly documented
Agent and channel boundaries
Several agents can share a memory space and learn from each other's cases. The management page does not describe rights per space.
RetrievalDocumented
Retrieval
Before the run, a recall step embeds the inputs and searches for similar cases; threshold, count and search type are configurable. Escalations use only the top match and apply the outcome without human review if it clears the threshold.
New conversationsNot publicly established
Session reset and compaction
The page describes no session logic or reset. According to the page only long text inputs are compacted before embedding. Traces with feedback or memory are exempt from TTL.
Cost and effort
5 plans · Prices as of 5 November 2026
Prices and included usage
Pricing checked
Free entry with Community, Basic from $25 per month, Standard and Enterprise on request. Agents consume Platform Units or Agent Units depending on the model.
| Plan | Price | Included |
|---|---|---|
| Community | Free ("Free forever") | Development draws from a monthly Community pool; the licensing page does not state its size. By default only your own models. Price source |
| Basic | From $25 per month | According to the page limited agent features, hosting in the European region. Agents can be designed but not run as processes. Price source |
| Standard | Sales; free 60-day trial | Build, test, run and share agents, own identity provider, hosting in a chosen region or on-premises. Price source |
| Enterprise | Sales | Additionally your own AI models, customer-managed keys and operation in several regions or on-premises. Price source |
| Usage (Unified Pricing) | 0.4 / 0.2 / 0.16 Platform Units per LLM call (Premium / Standard / Basic); 0.2 per run with your own model | UiPath-hosted models count in steps of 64,000 input tokens. On the Flex model, 2 / 1 / 0.8 Agent Units and 1 per run apply. Price source |
- The pricing page names unlimited agent calls for Standard and Enterprise and at the same time notes that usage can require additional purchases. The licensing page describes consumption-based billing.
- Conversational agents are billed per execution, meaning per agent turn.
- According to the pricing page, Basic is due to change.
- Third-party pricing was not used as evidence.
Setup
Platform account with a suitable plan, create an agent in Studio Web, publish it as a solution and deploy it. Escalations need Action apps deployed beforehand.
Operation & oversight
Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured.
Deployment covered: Cloud operation by the provider or self-operation with Automation Suite
Sources reviewed:
Editorial assessment
UiPath embeds agents in an existing operating environment with roles, traces, escalations and policies. That takes platform knowledge, and resuming individual runs stays open on the reviewed pages.
Show all 6 points
Evidence and open questions
25 sources · 2 open points
Open points
The reviewed sources contain no reliable information on these points. A gap says nothing about the product.
Requirements and limits: not yet confirmed
Sources and evidence
Agents documentation, licensing pages, pricing page, Automation Suite product page and region pages were reviewed. The docs mostly speak of Agents and Studio Web, product pages and memory docs of Agent Builder. Schedules and robot accounts are barely described on the reviewed agent pages. Example tasks are editorial. Own practical tests are pending.
Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.
Provider
UiPath offers a platform that according to the docs connects agents, robots and people in one automation system.
- Operator
- UiPath
- Organization
- Provider
- Website
- www.uipath.com
Changes to this profile 1
Agent capability of the UiPath platform: with Agent Builder in Studio you build agents with prompts, tools, context and human escalations and run them through Orchestrator.
These notes describe editorial changes. Source and pricing review dates appear in their respective sections.