Skip to main content

​
Changelog

Customer-facing updates to Candu Actions, grouped by release.
​
FixWizards
More reliable fullscreen wizard progress
Improved fullscreen progress handling for initial-submission wizard flows, so users can move from progress to the next wizard step without confusing stuck states.
​
Feature
Google Gemini support for Candu Actions

​
What changed

Candu Actions now supports Google-backed AI capabilities, including Gemini model options in the product UI and provider support in the backend.This release also includes Candu SDK 1.3.29, published alongside the Google provider work.

​
Why it matters

Teams can choose Gemini models for AI action workflows alongside existing provider options.
​
ImprovementWizards
Fullscreen action progress refined for multi-tool turns

​
What changed

Refined fullscreen tool progress for multi-action turns, including clearer phase coordination, smoother checklist motion, better pending states in simple chat, and Storybook coverage aligned to production behavior.

​
Why it matters

Longer-running action workflows are easier to follow while Candu Actions is working.
​
FeatureWizards
Attachment support for action workflows
Added attachment tool-call and dropzone support across the SDK, backend, and product UI, so wizards and action flows can collect files when a workflow needs documents, images, or other uploaded context.
​
Improvement
Agent version history and authoring polish
Added agent version history and shipped authoring improvements including quicker launch access, better tooltip positioning, copy support for tool slugs, and agent deletion support.
​
ImprovementWizards
Wizard interaction polish
Polished wizard interactions across the SDK, backend, and product UI after the primitives redesign, making guided action flows feel more consistent as users move between prompts, tool UI, and results.
​
FeatureWizards
Redesigned wizard UI primitives

​
What changed

Redesigned the wizard UI primitives across SDK, backend, and product surfaces, with updated interaction patterns and implementation support for richer guided flows.

​
Why it matters

Builders can create guided action flows from reusable primitives instead of one-off UI patterns.
​
FeatureWizards
More wizard form field types
Added number and email field support so wizard forms can collect structured input with fields that match the expected data.
​
ImprovementSecurity
Better context handling for long action histories

​
What changed

Added safeguards for wizard history context, including response-mask application, bounded truncation for oversized tool results, pruning for over-budget history, and logging when context preparation changes the payload.

​
Why it matters

Longer action sessions can keep useful context without carrying oversized tool output into later turns.
​
ImprovementWizards
Improved Candu Actions chat and progress UX

​
What changed

Improved simple chat, fullscreen progress, streaming behavior, tool timing, thinking states, and wizard tool UI for Candu Actions workflows.

​
Why it matters

Users get clearer feedback while tools run and smoother transitions back to the assistant response.
​
FeatureWizards
Thumbnail-grid choices for wizards
Added thumbnail-card support for wizard choice UI and aligned the backend configuration around the thumbnailGrid pattern, letting wizards present visual choices when text-only options are not enough.
​
Improvement
Support for multi-tool action turns
Added support for multiple tool responses in the SDK and updated backend guidance so Candu Actions can complete several related steps in a single turn.
​
Feature
Initial file upload support for action workflows
Added initial file upload support across the product UI, SDK, and backend, including S3-backed upload handling for workflows that need uploaded files as user input.
​
Fix
Clearer handling for invalid tool calls
Added handling for cases where the model asks for a tool that is not available, marking those situations as errors so failed tool calls are easier to detect, recover from, and debug.
​
FeatureWizards
Modal and inline wizard launch modes

​
What changed

Added modal wizard mode, product-side wizard open mode, initial-submission triggering for agent actions, and an inline selector trigger.

​
Why it matters

Teams can start guided action flows from more places in the product, with launch behavior that better matches the surrounding UI.
​
Improvement
Improved action logs and session visibility

​
What changed

Improved action logs, session detail screens, and backend log handling.

​
Why it matters

Teams have better visibility into what happened during an action session when they need to review or troubleshoot it.
​
FeatureWizards
Simple mode and visible tool progress

​
What changed

Added simple mode for Candu Actions experiences and surfaced tool progress while agents are working.

​
Why it matters

Product teams can embed lighter-weight AI action experiences while still showing users when work is in progress.
​
Feature
Agent placement configuration

​
What changed

Added placement data models, backend storage, controller support, and product UI for configuring where Candu Actions agents appear.

​
Why it matters

Teams can manage where action experiences show up instead of treating placement as a one-off implementation detail.
​
Improvement
Code Chat editing polish

​
What changed

Added multi-line input and copy button support for Code Chat, plus better defaults for snapshot capture.

​
Why it matters

Code Chat helps teams talk directly to their product and discover possible actions; these editing improvements make longer prompts and repeated iteration easier.
​
Improvement
Unified SDK foundation for Candu Actions

​
What changed

Integrated the Candu Actions experience into the SDK and moved the product toward a unified SDK path.

​
Why it matters

A shared SDK foundation makes it easier to ship Candu Actions consistently across embedded and product-managed surfaces.
​
Improvement
Session summaries for agent activity

​
What changed

Added support for generating session summaries from agent activity and applying those summaries at the right points in the workflow.

​
Why it matters

Longer sessions become easier to review without reading every intermediate turn.