Governed Agent Execution
July is a major step forward for Governed Agents in Clear Ideas.
Agents can now handle more complete business processes with better run visibility, stronger output handling, richer input files, controlled external access, and more reliable execution across longer agent runs. They can also run with Private Models and Zero Data Retention AI Models where the selected model path supports that handling, helping teams match sensitive recurring work to the right execution environment.
What's new
- More capable agent runs with improved handling for multi-step agent processes, generated files, source files, and run outputs
- Private model execution options for selected agent runs that need tighter model-path controls
- Controlled tools and external access so agent capabilities can follow approved source, connector, and organization policy rules
- Loop steps with goal-based execution so agents can repeat a task until a defined goal or output condition is met
- Input files for agent instructions so agents can work with selected files, file references, and dynamic expressions as part of a governed process
- Structured values in agent runs with clearer formatting expectations for complex variables and reusable data
- More consistent run records so agent activity, outputs, and generated files are easier to review after completion
- Better failure messages and recovery behavior when a run cannot complete as expected
Why it matters
Many valuable business processes are not a single prompt. They involve source files, intermediate outputs, repeated checks, approvals, and handoffs.
These changes make Governed Agents more useful for work such as diligence review, reporting, research briefs, policy checks, recurring summaries, and other repeatable processes where teams need both automation and a clear record of what happened. They also help administrators give agents useful capabilities without opening broad, unmanaged access to external resources.
Agent Designer and Builder Improvements
Clear Ideas now gives teams a smoother path from idea to working agent.
The agent designer can capture more context about how an agent was created, show clearer metadata about the design process, and support more practical editing after the first draft. Teams can also duplicate existing agents, refine saved prompts, and manage agent details with less manual rebuilding.
What's new
- Richer Agent Designer metadata showing useful creation context, selected model information, and clarifying questions
- Improved agent duplication so teams can copy an existing agent and adapt it for a new use case
- Better saved prompt handling across agent editing
- Cleaner prompt authoring tools for models, variables, source files, output settings, and advanced options
- Improved schedule editing for recurring agent runs
- More consistent agent lists, avatars, filters, and run views across the agent workspace
Why it matters
The first version of an agent is rarely the final version. Teams need to experiment, copy, adapt, and review agents without losing the context behind them.
These updates make agent creation feel more like an ongoing operational process: design the first draft, inspect how it was built, duplicate it when a similar process appears, and keep improving the system from real usage.
Private Models and Zero Data Retention AI Models
July introduces Private Models in Clear Ideas.
Teams can now choose private execution paths for selected AI Chat and Agent runs, alongside Zero Data Retention AI Models where the selected provider and model path support that handling. These options help organizations match each task to the right balance of capability, sensitivity, and policy control.
What's new
- Private Models for selected AI Chat and Agent runs that need private execution
- Zero Data Retention AI Models available where the selected model path supports Zero Data Retention handling
- Improved model selection across chat and agent experiences so users can better understand which model they are using
- Better support for multimodal models when prompts include images or file context
- User-specific model visibility so model choices can be tailored by account and policy
- Updated model metadata for administrators managing available options
Why it matters
Different work calls for different model paths. Some teams need speed, some need reasoning depth, some need image support, and some need private execution or Zero Data Retention handling for sensitive work.
These updates make model choice more understandable for users and more manageable for administrators. They also make it easier for organizations to apply policy-aware AI without treating every task, model, and data sensitivity level the same way.
Content Management and Indexing Controls
Clear Ideas now gives administrators more control over how content is selected, indexed, extracted, and managed.
Folders can be marked for indexing exclusion, selectable content lists are easier to sort and paginate, archive handling has been improved, and file extraction actions now provide clearer confirmation before work begins.
What's new
- Folder indexing controls for excluding selected folders from AI indexing
- Clearer folder properties with localized messages and better routing from content menus
- Improved archive handling for extracting supported compressed file types
- Confirmation prompts for extraction actions so users understand when files will be unpacked or processed
- Better selectable content lists with sorting, pagination, and refined bulk selection
- More reliable document identity handling during extraction and post-upload processing
Why it matters
Governed AI depends on well-managed source content. Administrators need to decide what should be searchable, what should be excluded, and how uploaded materials should be prepared before users rely on them.
These changes make day-to-day content administration more predictable while reducing friction for larger workspaces.
Redaction Delivery Controls
Redaction now includes clearer delivery controls for role-based access.
Administrators can configure which user roles receive redacted PDFs, and related file, site, and question views have been refined so the redaction experience is easier to understand.
What's new
- Role-based redacted PDF delivery controls
- Improved PDF redaction settings in file and site configuration views
- Cleaner redaction and watermarking layouts across responsive screens
- Better integration with questions and document review flows
Why it matters
Redaction is not only about removing sensitive information. It is also about ensuring the right people receive the right version of a document.
Role-based delivery controls help teams standardize that decision inside Clear Ideas instead of relying on manual distribution steps outside the workspace.
Supporting Improvements
July also includes ongoing reliability, security, localization, dependency, and performance work across the Clear Ideas web app and API.
Several improvements in this release are intentionally described at a high level because they strengthen Clear Ideas without changing the product surface customers need to understand. Together, they make Governed Agents, Private Models, Zero Data Retention AI Models, content management, redaction, and administration more reliable for teams using Clear Ideas in sensitive collaboration.