Release Notes

A catalog of the releases most relevant to UiPath Python SDK users (breaking changes and notable updates). Full details live in each GitHub release, linked below.

uipath (SDK & CLI)

Release	Date	What's relevant	Notes
v2.12.0	2026-06-30	Eval framework: AgentExecution → WorkloadExecution; fields agent_trace/agent_output → workload_trace/workload_output; evaluate()/validate_and_evaluate_criteria() keyword agent_execution → workload_execution	🚨 Breaking — see migration
v2.10.0	2026-02-27	Coded function schema type changed from "agent" to "function"	🚨 Breaking
v2.9.0	2026-02-23	platform extracted to uipath-platform, context grounding contract changes, uipath dev defaults to web	🚨 Breaking
v2.2.0	2025-11-26	Python 3.11+ required, UiPath import moved to uipath.platform, configuration architecture redesign	🚨 Breaking

uipath-langchain

Release	Date	What's relevant	Notes
v0.10.0	2026-04-23	Transport/auth split into new uipath-llm-client and uipath-langchain-client packages (legacy preserved)	Non-breaking

uipath-runtime

Release	Date	What's relevant	Notes
v0.3.0	2025-12-18	UiPathDebugBridgeProtocol renamed to UiPathDebugProtocol	🚨 Breaking (protocol implementers only)

Migration: v2.12.0 (eval WorkloadExecution rename)

The eval framework's central execution type and its keyword parameters were renamed to unified-evals terminology. The value passed to every evaluator represents a workload execution (agent, process/orchestration, case management, …), not specifically an agent.

These are breaking changes for code that uses the affected public names. Update as follows:

Before	After
from uipath.eval.models import AgentExecution	from uipath.eval.models import WorkloadExecution
WorkloadExecution(agent_trace=...)	WorkloadExecution(workload_trace=...)
WorkloadExecution(agent_output=...)	WorkloadExecution(workload_output=...)
execution.agent_trace / execution.agent_output	execution.workload_trace / execution.workload_output
evaluator.evaluate(agent_execution=...)	evaluator.evaluate(workload_execution=...)
evaluator.validate_and_evaluate_criteria(agent_execution=...)	evaluator.validate_and_evaluate_criteria(workload_execution=...)

What still works:

• AgentExecution (the class name) is soft-deprecated: importing it still resolves to WorkloadExecution and emits a DeprecationWarning (removed in uipath 3.0). Note the fields are not aliased — constructing AgentExecution(agent_output=..., agent_trace=...) now raises a ValidationError; rename the fields as above.
• Custom evaluators that override evaluate / validate_and_evaluate_criteria with the old agent_execution parameter name keep running, because the runtime dispatches positionally. Only callers passing agent_execution= by keyword are affected.

Unchanged (intentionally): the WorkloadExecution.agent_input and expected_agent_behavior fields, and the agent_output / agent_execution_time fields on EvalRunUpdatedEvent and StudioWebProgressItem (the latter a Studio Web wire contract).