Agents release notes

Last updated Feb 19, 2026

February 2026

February 18, 2026

Model deprecation: Claude Sonnet 3.5

Anthropic has announced the deprecation of Claude Sonnet 3.5. Effective February 27, 2026, Sonnet 3.5 will no longer be available as a selectable model in Agent Builder across all environments. Agents currently deployed using Sonnet 3.5 will automatically transition to Sonnet 4.5. This automatic fallback ensures there is no disruption to existing deployments or business operations.

February 11, 2026

New just-in-time context and built-in tools for file-based agent workflows

This release introduces just-in-time (JIT) context and two powerful built-in tools, DeepRAG and Batch Transform, that make it easier to ground agents in files, process large datasets, and evaluate results at scale. Together, these capabilities reduce setup overhead, improve traceability, and enable richer file-centric agent workflows.

Just-in-time context for agents

Agents can now consume files and documents as just-in-time context, passed directly at runtime rather than pre-indexed or pre-configured.

With JIT context:

Files are provided as agent inputs and resolved automatically at execution time.
No storage buckets, indexes, or upfront ingestion steps are required.
Agents can operate on dynamic, user-provided content with minimal configuration.

This model simplifies workflows where input data changes frequently or is only available at runtime.

Batch Transform tool

The Batch Transform tool enables agents to process and enrich structured files, such as CSVs, at scale using AI.

Key capabilities:

Apply the same transformation logic to every row in a file.
Generate new columns or rows based on model reasoning and user-defined rules.
Pass files directly as agent inputs and receive transformed files as outputs.
Inspect input and output files directly from the execution trace.

Batch Transform is also integrated with agent evaluations, with support for line-by-line evaluation of structured outputs and a new CSV Column Exact Match evaluator.

DeepRAG tool

The DeepRAG tool enables agents to analyze and synthesize information across large documents using retrieval-augmented generation, without requiring a long-running or prebuilt index.

Key capabilities:

Analyze documents of up to approximately 1,000 pages.
Perform semantic retrieval across long-form content such as PDFs or text files.
Generate comprehensive, synthesized answers grounded in source material.
Include citations that reference the original documents.
Inspect inputs, outputs, and citations directly from the execution trace.

DeepRAG is well suited for research, policy analysis, knowledge synthesis, and other scenarios that require deep understanding of complex documents with traceability back to sources.

Introducing data residency in Switzerland for Agents

As part of our ongoing cloud expansion, we’re bringing complete data sovereignty to the Switzerland region for Agents. This ensures your data is securely stored and replicated across multiple availability zones within Switzerland, offering full control and compliance confidence.

For details, refer to Data residency in the Automation Cloud Admin guide.

Pricing updates

We are simplifying Agents pricing to make it more predictable and developer-friendly. Customers who bring their own LLM are now charged once per agent run instead of per individual LLM call, with a built-in 30% discount. Pricing for UiPath-hosted models remains unchanged, and tool usage continues to be charged consistently, regardless of the model selected.

We’re also making agent development effectively free. Pro Users, App Test Developers, and Automation Developers now receive 2,500 LLM calls per day to support agent design, testing, and evaluation, so teams can iterate faster without worrying about development costs.

For details, refer to Licensing.

Agent escalation memory

Agent escalation memory, previously available in preview, will be temporarily disabled while we work on an improved experience aligned with episodic memory updates in Studio Web and Agents Instance Management. The updated experience is expected to return in the coming weeks.