SaaS Release — 2026-08-17 to 2026-08-30
Window: 2026-08-17 → 2026-08-30
Highlights
Section titled “Highlights”AI spend in dollars
Section titled “AI spend in dollars”Admins can review model spend in dollars alongside credit usage. The AI Spend dashboard includes cost-aware analytics so teams can understand real inference spend, including calls that use customer-provided model access.
More reliable tool access for large integration sets
Section titled “More reliable tool access for large integration sets”Agents with many available tools use a more scalable tool-discovery path instead of sending an oversized tool list to the model. This improves reliability for agents connected to broad integrations and reduces cases where an important tool disappears from a run. The unused MCP Prompts surface has also been removed so connected clients focus on supported tool operations.
Integrations
Section titled “Integrations”Microsoft government cloud and Entra ID coverage
Section titled “Microsoft government cloud and Entra ID coverage”Microsoft integrations support more government-cloud tenant scenarios, including Azure Government and GCC High variants across Microsoft cloud services. Entra ID coverage also expands with reporting tools for application activity, credential usage, tenant recommendations, and SLA status.
Expanded security and identity integrations
Section titled “Expanded security and identity integrations”Kindo adds broader cloud, identity, and security-operation coverage across key integrations:
- AWS and Google Cloud security tools for cloud-asset investigation and action.
- SailPoint task-status, result-retrieval, and policy-governed operational tools.
Improvements
Section titled “Improvements”Stronger admin governance and self-service troubleshooting
Section titled “Stronger admin governance and self-service troubleshooting”Organization settings are easier to navigate: Model Access and DLP, Integration and Tool controls, Audit Log, Sharing Permissions, and Agent Telemetry Export appear as first-level Organization settings instead of being nested under a Security group. Audit logs also capture integration connection and deletion events, and the public MCP gateway applies organization policy and DLP controls before returning tool results. Agent run pages also show the Inspect view and identify the failing step more clearly when a run fails.
More model and API control
Section titled “More model and API control”DeepSeek V4 Pro is updated to the 0813 revision, with a larger context window and lower listed provider pricing. DeepHat V3 can be enabled as a centrally managed model option, direct DeepHat API usage can be tracked with account-level credentials, and chat-completion API calls honor JSON response-format requests more consistently.
Public setup and API documentation
Section titled “Public setup and API documentation”Kindo’s SSO setup guides for Microsoft Entra ID, Okta, and Google Workspace are hosted directly in the docs site. The Conversations API documentation also clarifies that item-reading endpoints apply to any authorized conversation, not only conversations created through the Responses API. Agent documentation also covers more agent types, scheduled runs, direct webhook triggers, programmatic execution, and best-practice cross-links.
Reliability fixes for agent work, auth, uploads, and ingestion
Section titled “Reliability fixes for agent work, auth, uploads, and ingestion”This release includes several customer-observable fixes:
- Long-running agent conversations are less likely to dead-end after a tool call, and persistent workspaces are less likely to lose artifacts during multi-session work.
- Chat persistence handles transaction conflicts more gracefully instead of failing the run.
- Email sign-in code delivery, invited-user sign-in, and knowledge-base file uploads are more reliable.
- Oversized API requests return a clear
413 Payload Too Largeresponse instead of an opaque server error. - File ingestion handles provider-specific embedding limits more safely.
- Integration tool discovery recovers from rollout regressions that previously made some integration capabilities appear unavailable.