SaaS Release — 2026-09-28 to 2026-10-11
Window: 2026-09-28 → 2026-10-11
Highlights
Section titled “Highlights”Create agents from the public API
Section titled “Create agents from the public API”You can create an agent by sending an agent definition to the Kindo public API. This gives teams a scriptable path for creating agents from exported definitions or generated configuration, with clear validation errors for malformed definitions and permission errors when agent creation is not allowed. See Create an agent to get started.
Persistent workspaces are generally available
Section titled “Persistent workspaces are generally available”Persistent workspaces are out of beta in the agent builder. Agents can keep using the same workspace across runs.
Public integration tool catalog
Section titled “Public integration tool catalog”The docs now include a tool catalog (beta) for Kindo integrations. Admins and builders can see which integrations, connection variants, authentication methods, and tools are available without needing to inspect the product internals. Because the catalog is a point-in-time snapshot, an entry can differ from what your organization sees in product.
Integrations
Section titled “Integrations”Expanded AI, security, and compliance integrations
Section titled “Expanded AI, security, and compliance integrations”Google SecOps adds detection-engineering tools. Kindo also expands its early-access integration coverage across AI governance, security operations, compliance, and operational data with Azure AI Foundry, Vanta, Aikido Security, and Airtable; ask your Kindo representative to enable them for your organization.
Broader governed access for security tools
Section titled “Broader governed access for security tools”Several security integrations gained more complete connection and read capabilities:
- Abnormal Security supports read-only, audited message-content and attachment retrieval for approved agents.
- Microsoft Defender for Endpoint can connect with application permissions through a customer-managed app registration; contact your Kindo representative for setup guidance.
- Microsoft Teams, OneDrive, and Outlook Calendar can support application-permission deployments alongside delegated sign-in; contact your Kindo representative if your tenant requires application permissions.
- CyberArk Privilege Cloud connection setup uses the credential flow CyberArk accepts.
More reliable core integrations
Section titled “More reliable core integrations”Box, Grafana, SailPoint, and Tenable Vulnerability Management run more reliably, with no change to their existing tools.
Improvements
Section titled “Improvements”Larger sandbox file sharing
Section titled “Larger sandbox file sharing”Agents can share large sandbox-generated files with users, without the previous small-file cap.
Clearer tool approvals and chat file downloads
Section titled “Clearer tool approvals and chat file downloads”Tool approval cards wrap long inputs so they’re easier to review, and you can turn this off with Wrap lines. File chips in chat also make it clearer that selecting one downloads the file.
More accurate model usage and billing views
Section titled “More accurate model usage and billing views”AI spend and usage reporting handles more pricing edge cases, including prompt-cache reads on deployments missing a cache-read rate and models with missing per-token rates. Command Center usage views are more reliable when comparing model usage across an organization.
Stronger content-access checks
Section titled “Stronger content-access checks”Content-access checks now treat explicit denials as denials and filter out removed or deleted content. This keeps agent and tool access aligned with the content lifecycle and authorization rules admins expect.
More reliable model selection
Section titled “More reliable model selection”A selected chat model now displays correctly, and replacing a globally managed model preserves a valid replacement instead of clearing an organization’s default chat model.
Clearer schedule-generation errors
Section titled “Clearer schedule-generation errors”When you use Generate Schedule with AI on an agent’s schedule, Kindo now explains why a request cannot be expressed as a single schedule instead of showing a generic processing error.