# UiPath Agents 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. Canonical profile: https://ki-agentenkompass.de/en/agents/uipath/ Editor: [Michael Meierhoff](https://ki-agentenkompass.de/en/about/) Sources reviewed: 2026-11-05 Basis: provider documentation. Examples are editorial suggestions, not hands-on product tests. ## Overview 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. - Provider: UiPath - Product family: UiPath Agents - Profile type: Agent capability of an automation platform - Environment: Automation Cloud, Dedicated and Test Cloud; self-operated through Automation Suite. Design in Studio Web. - Connections: Tool types according to the docs: Activities, agents, API workflows, automations, Functions, IXP, MCP servers and client-side tools. Models from OpenAI, Anthropic and Google or your own. - 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. - Pricing model: Community free, Basic from $25 per month, Standard and Enterprise through sales. Agent runs also consume Platform Units or Agent Units; the licensing page gives no money values for them. ## Key features ### 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: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/building-an-agent-in-studio-web ### Tools An agent only acts through tools it is explicitly given access to, such as automations, API workflows, other agents or MCP servers. Source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-tools ### Escalations The agent hands decisions to people through Action apps in Action Center. For each outcome you set whether it continues or ends. Source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-escalations ### Guardrails Built-in checks for personal data, prompt injection and harmful content can log, block or escalate. They depend on the licence plan. Source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/out-of-the-box-guardrails ### Traces and evaluations Traces show the steps, inputs, outputs, errors and token usage of a run. Evaluation sets test agents against defined inputs. Source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-traces ### Agent memory (preview) Episodic memory recalls similar earlier cases, escalation memory reapplies earlier human decisions. Source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-memory ### Coded Agents Agents developed outside UiPath can be deployed on the platform and have their own licensing model. Source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/coded-agents-licensing ### Self-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: https://www.uipath.com/product/automation-suite ## Example workflows ### 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. ## Requirements and limits ### Running from Standard Requirement: On Unified Pricing, Basic plans can only design agents. They run as processes from Standard Trial, Standard and Enterprise. Source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/licensing ### Not for every workflow Limitation: The docs name high-risk financial transactions, zero-error regulatory workflows and multi-step data processing without deterministic safeguards as poor fits. Source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/about-uipath-agents ### Differences by delivery Limitation: Dedicated offers no conversational agents and no agent memory and only part of the guardrails. Source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agents-feature-availability ### Not in Public Sector Limitation: The product overview lists Agents as available for Automation Cloud, Dedicated and Automation Suite, not for Public Sector. Source: https://docs.uipath.com/overview/other/latest/overview/product-and-feature-availability-across-delivery-options ### Automation Suite on Linux Not yet confirmed: The product page names agent capabilities on Linux as planned for October 2026. Whether they have shipped was not confirmed. Source: https://www.uipath.com/product/automation-suite ### Retrying failed runs Not yet confirmed: The page on running names the failed status but describes no retry or recovery. Source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/running-agents ## 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. ## 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 Documented · Sources accessed on: 2026-11-05 - [UiPath: AI features and model routing](https://docs.uipath.com/automation-cloud/automation-cloud/latest/admin-guide/ai-features-and-model-routing) - [UiPath: Connecting open-source models](https://docs.uipath.com/automation-cloud/automation-cloud/latest/admin-guide/connecting-oss-models-aitl) - [UiPath: Bring your own model](https://www.uipath.com/blog/product-and-updates/introducing-bring-your-own-model-to-uipath) - [UiPath Automation Suite: LLM support](https://docs.uipath.com/automation-suite/automation-suite/2.2510/admin-guide/about-llm-integration) ## 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: 2026-11-05 ### Data residency by plan With requirements 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 covered: EU Documented reason: The page governs platform data and does not name Agents separately. Official source: https://docs.uipath.com/automation-cloud/automation-cloud/latest/admin-guide/data-residency-automation-cloud ### Model routing for agents With requirements 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 covered: EU Documented reason: Depends on the chosen model and the policy. Official source: https://docs.uipath.com/automation-cloud/automation-cloud/latest/admin-guide/ai-features-and-model-routing Official source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agents-governance ### Switzerland and United Kingdom With requirements 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 covered: Switzerland and United Kingdom Documented reason: Depends on the model; the data residency page does not name either region. Official source: https://docs.uipath.com/automation-cloud/automation-cloud/latest/admin-guide/ai-features-and-model-routing ### Prompt injection check in the US With requirements The guardrail runs at Noma Security in the US, so data can be processed outside the tenant region. Region covered: EU/EEA, United Kingdom and Switzerland Documented reason: Service location named by the provider. Official source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/out-of-the-box-guardrails ### Self-operation Available With Automation Suite, configurations, data and audit trails stay in your Kubernetes environment according to the post. Models can run there as well. Region covered: Location chosen by the operator Documented reason: The post generally names sovereignty and in-country processing, but no individual countries. Official source: https://www.uipath.com/blog/product-and-updates/agentic-ai-now-where-your-data-lives Official source: https://www.uipath.com/product/automation-suite ## 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: 2026-11-05 ### Account used Documented with conditions: From Standard, an own identity provider is possible according to the pricing page. The page on running does not say which identity or robot account an agent runs under. Evidence: https://www.uipath.com/pricing Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/running-agents ### Admin approval Documented: Escalations lead through Action apps to named people or groups. A Studio Web policy can require a human escalation before publishing. Guardrails can also escalate. Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-escalations Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agents-governance Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/out-of-the-box-guardrails ### Data access & team sharing Documented with conditions: 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: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-tools Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agents-governance ## Prices and included usage Pricing checked: 2026-11-05 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. ### 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: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/licensing ### 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: https://www.uipath.com/pricing ### 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: https://www.uipath.com/pricing ### Enterprise Sales Additionally your own AI models, customer-managed keys and operation in several regions or on-premises. Price source: https://www.uipath.com/pricing ### 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: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/licensing 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. Pricing details: https://www.uipath.com/pricing ## 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: 2026-11-05 ### Files and working state Documented: According to the docs agents run in a secure serverless environment. Traces are kept for 90 days by default, with TTL enabled for 1 or 7 days. In Automation Suite, configurations and audit trails sit in your Kubernetes environment. Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/running-agents Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-traces Evidence: https://www.uipath.com/blog/product-and-updates/agentic-ai-now-where-your-data-lives ### Sign-ins Documented with conditions: From Standard, an own identity provider is possible. The reviewed agent pages do not say which account an agent uses in target systems. Evidence: https://www.uipath.com/pricing Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/running-agents ### Starting further runs Documented: After publishing as a solution the agent runs as an Orchestrator job, started via API, Maestro node or process activity. The traces docs mention scheduled runs. The docs name triggers from system events or user input. Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/running-agents Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-traces Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/about-uipath-agents Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/publish-the-agent ### Human intervention Documented: Long steps such as human actions can pause the run, and it then continues automatically. Escalations go through Action Center; for each outcome you choose Continue or End. Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/running-agents Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-escalations ### Detecting errors Documented: Traces mark steps as successful, failed or retried and show the error. The Agents dashboard reports errors and falling scores. To view them you need Logs.View and Jobs.View. Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-traces Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/managing-uipath-agents ### Resuming after failures Documented with conditions: The management page names error handling and recovery through BPMN events for agent instances in business processes. For individual agent runs, the running page describes no retry. Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/managing-uipath-agents Evidence: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/running-agents ### 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. ## Memory and conversation continuity Scope: Agent memory as a preview, in shared memory spaces Sources accessed: 2026-11-05 Episodic and escalation memory in memory spaces that can be reviewed and edited. Preview, not on Dedicated; retention period not stated. ### Storage and scope Partly documented: 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. - [UiPath Docs: Agent memory](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-memory) - [UiPath Docs: Managing memory spaces](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/instance-management-memory-spaces) - [UiPath Docs: Agents feature availability](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agents-feature-availability) ### Selection and timing Documented: 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. - [UiPath Docs: Agent memory](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-memory) - [UiPath Docs: Agent traces](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-traces) ### Inspect, edit and delete Documented: 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. - [UiPath Docs: Agent memory](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-memory) ### Agent and channel boundaries Partly documented: Several agents can share a memory space and learn from each other's cases. The management page does not describe rights per space. - [UiPath Docs: Agent memory](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-memory) - [UiPath Docs: Managing memory spaces](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/instance-management-memory-spaces) ### Retrieval Documented: 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. - [UiPath Docs: Agent memory](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-memory) ### Session reset and compaction Not publicly established: 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. - [UiPath Docs: Agent memory](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-memory) - [UiPath Docs: Agent traces](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-traces) ## Context The docs advise against high-risk financial transactions and zero-error regulatory workflows. Running agents starts at Standard. Agent memory is in preview and missing on Dedicated. List prices for Standard and Enterprise are not public. 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. ## Open points Requirements and limits, not yet confirmed: - Automation Suite on Linux - Retrying failed runs ## Sources - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/about-this-guide - https://docs.uipath.com/overview/other/latest/overview/product-and-feature-availability-across-delivery-options - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/conversational-agents-licensing - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/about-agents - https://www.uipath.com/platform/agentic-automation - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/building-an-agent-in-studio-web - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-tools - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-escalations - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/out-of-the-box-guardrails - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-traces - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-memory - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/coded-agents-licensing - https://www.uipath.com/product/automation-suite - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/licensing - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/about-uipath-agents - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agents-feature-availability - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/running-agents - https://www.uipath.com/pricing - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agents-governance - https://www.uipath.com/blog/product-and-updates/agentic-ai-now-where-your-data-lives - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/publish-the-agent - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/managing-uipath-agents - https://docs.uipath.com/automation-cloud/automation-cloud/latest/admin-guide/data-residency-automation-cloud - https://docs.uipath.com/automation-cloud/automation-cloud/latest/admin-guide/ai-features-and-model-routing - https://docs.uipath.com/agents/automation-cloud/latest/user-guide/instance-management-memory-spaces ## Provider UiPath offers a platform that according to the docs connects agents, robots and people in one automation system. - Operator: UiPath - Organization: Provider - Website: https://www.uipath.com - Reviewed: 2026-11-05 - Provider source: https://docs.uipath.com/agents/automation-cloud/latest/user-guide/about-agents - Provider source: https://www.uipath.com/platform/agentic-automation [Sources and method](https://ki-agentenkompass.de/en/sources/)