> ## Documentation Index > Fetch the complete documentation index at: https://docs.genesiscomputing.com/llms.txt > Use this file to discover all available pages before exploring further. # v1.19.42 > Genesis Data Agents v1.19.42 release notes ## Genesis v1.19.42 — May 15, 2026 ### Azure DevOps and Microsoft Teams integrations, Mission Debrief, scoped connection defaults, Context Graph Google Drive scanner, and broad reliability improvements across Git, chat, and compression #### Azure DevOps Integration Genesis agents can now connect directly to Azure DevOps without a third-party bridge. * Agents can query Boards work items, inspect and manage Repos pull requests, trigger Pipelines runs, and poll per-environment deployment status — all through a single PAT-backed connection per ADO organization. * Classic Release Definitions are fully supported: agents can discover available release pipelines, trigger releases, and track per-environment deployment outcomes. * The Azure DevOps connection card in the dashboard is now branded with the correct logo and color, making it easy to identify at a glance. * GitLab push rejections caused by server-side pre-receive hooks (signed-commit requirements, GPG enforcement) are now recovered automatically by replaying the change through the Commits API, eliminating a class of push failures on protected repositories. #### Microsoft Teams Integration Genesis agents can now send messages and notifications to Microsoft Teams. * Agents can post messages to Teams channels and 1:1 chats through Microsoft Graph delegated OAuth, with no manual credential plumbing beyond the initial connection setup. * Outbound webhook notifications include bounded-retry delivery assurance, so P0 alerts are reliably delivered even when Teams is briefly slow. * Agents can author structured Adaptive Cards — including the five most common alert and summary shapes — without composing raw JSON. * 1:1 direct message chat creation now correctly resolves recipient identity, fixing delivery failures that occurred when initiating new 1:1 conversations. * Public documentation for Microsoft Teams Webhook configuration is now available. #### Mission Debrief Agents can now generate, store, and display structured debrief reports at the conclusion of a mission. * A Mission Debrief captures the outcomes, decisions, and key artifacts produced during a multi-step agent engagement and renders them as a structured, versioned document. * Debrief reports are accessible from the dashboard and update automatically when an agent completes a mission phase. * Bedrock streaming reliability was hardened during debrief development: the 600-second ceiling bug that caused long-running Bedrock calls to terminate mid-sentence is now fixed, and per-call streaming configuration is now isolated to prevent one call's settings from affecting another. * The tool-output governor default was reduced from 96 KB to 32 KB, preventing runaway context growth from large tool responses. #### Scoped Connection Defaults Users and administrators can now declare a preferred connection of any type without modifying system-wide configuration. * A three-tier default system (user → role → system) lets individuals set their own preferred database, Git, or cloud connection. When no user-level default is set, the role-level default applies; system defaults serve as the final fallback. * Connection defaults are configurable from the dashboard and respected automatically by agents when selecting which connection to use for a task. * Ambiguous role defaults now surface a visible warning in the UI rather than silently picking an arbitrary connection. #### Context Graph: Google Drive Scanner The Context Graph can now ingest unstructured knowledge stored in Google Drive. * A new Google Drive scanner reads documents from connected Drive folders, clusters them into logical wiki pages by folder structure, generates summaries, and links the content to related data assets already in the graph. * Drive content becomes available to agents through the same wiki retrieval machinery that serves code, SQL, and BI assets, so agents can answer questions that span structured warehouse data and unstructured documents in a single conversation. #### RBAC: Granular PAT Connection Permissions and App-Tag Role Grants Access control gained two new capabilities for administrators managing large teams. * PAT connection permissions are now split into five category-scoped grants — databases, Git, cloud, pipelines, and tools — so administrators can selectively allow PAT-backed connections per category rather than granting or denying all at once. Existing roles are migrated automatically. * The role editor's Apps tab now supports bulk access-level assignment by app tag: operators can grant a permission level to all Genesis Apps sharing a given tag in a single action, rather than configuring each app individually. #### Git: Reliability and Author Identity Several reliability and correctness improvements were made to Git-backed operations. * Stale content reads from managed-remote repositories are now prevented by a comprehensive fix that covers all reader types — document parsers, search, list, and find operations — not just the explicit-branch path that was patched previously. * Custom commit author identity is now supported for GitHub, GitLab, and Bitbucket connections: administrators can configure a display name and email for Genesis-generated commits, satisfying auditor requirements for human attribution. * Co-author trailers can now be added to any commit, providing proper attribution when a human and an agent collaborate on a change. * The API-commit fallback path now correctly diffs against the remote branch head on the second and subsequent edits to an open branch, preventing silent content loss from stale base comparisons. * Git author name validation now handles edge cases that previously caused commit failures. * Git Browser search results now display the correct directory hierarchy rather than showing matched files as flat entries directly under their repository root. #### Chat and Streaming Improvements The chat interface received several fixes that eliminate visible artifacts and improve the real-time experience. * Streaming text no longer flickers or disappears mid-turn; the race condition causing this was eliminated along with a stale-closure WebSocket handler. * Internal tool-envelope rows, reasoning markers, and checkpoint entries no longer appear in the user-visible transcript during streaming or in completed threads. * Duplicate user messages that appeared optimistically and then again on server confirmation are now correctly collapsed into a single entry. #### Thread Compression Long-thread compression is more reliable and now shows progress in the UI. * Compression now runs in parallel — LLM summaries for multiple conversation segments are generated simultaneously, reducing observed compression time significantly. * A live progress indicator in the chat panel shows compression status as it runs, replacing the previous silent wait. * A race condition where the main process could start a competing in-process compression while the background worker was still running is now fixed, eliminating the dominant failure mode for large-thread compression. * The manual compression panel now immediately refreshes context statistics and marks the last-prompt snapshot as stale after compression completes, ensuring the displayed counts are accurate. #### Connections and AI Providers * AWS Bedrock connections can now be pointed at enterprise proxy endpoints (custom hostname, API route, extra HTTP headers, TLS verification toggle), enabling deployments behind API Gateway fronts. * Snowflake account identifier inputs are now normalized automatically, preventing connection failures caused by formatting differences between account input styles. * Demo database connections (baseball, Formula 1, HCLS, etc.) can now be hidden by administrators from the Connections list without triggering the seed path that would immediately recreate them. All demo connections can be bulk-hidden in one action. * Hidden demo connections no longer appear in connection-type maps, bot-to-connection allowlists, or Cortex Code CLI candidate lists. * LLM model context-window caps, fallback lists, and temperature-support configuration are now maintained in a central configuration file rather than scattered through Python source, making custom model configurations easier to manage. * Connection-backed provider APIs are now guarded against bypass: agents cannot route provider API calls through the generic web tool using manually collected tokens when a configured connection already covers that provider. * Cortex REST embedding on Snowflake SPCS deployments now correctly falls back to the SQL-based embedding path when the REST endpoint returns a 403, fixing document indexing and thread search failures on SPCS. #### MCP Gateway * The MCP gateway v2 is now fully compatible with both the previous and current SPCS bridge versions, closing gaps in diagnostic actions, parameter forwarding, and long-poll handling. * The Claude Code setup panel now provides clearer guidance for OAuth client configuration. #### Observability * Genesis can now forward metrics to any standard OTLP-compatible collector (Prometheus, Grafana, Datadog) via an opt-in push path, running in parallel with the existing SQL metrics pipeline. * LLM call logs now include the stream finish reason and configured output token cap at the end of every call, making it straightforward to distinguish natural model stops from token-budget truncations during triage. * Public documentation for OTLP metrics export and distributed tracing configuration is now available. #### Helm / Kubernetes Deployment * Kubernetes deployments in corporate or restricted-network environments can now automatically fetch custom CA certificates from AWS Secrets Manager or GCP Secret Manager at pod startup and merge them into the container trust store, eliminating manual certificate injection steps. * The Genesis Helm chart is now published to Docker Hub and version-bumped automatically on each GA release. * IaC-friendly configuration is now supported: operators can supply a single YAML config file covering LLM provider credentials, database connections, and application settings, with secret values resolved at runtime from AWS Secrets Manager or GCP Secret Manager. #### Support * A "File support request" action is now available in the chat thread menu. Users can describe what they expected and what they observed; the dialog posts a structured prompt into the active thread to initiate support bundle collection. #### Jira Integration * Jira auto-pickup is now enabled by default for connected Jira workspaces. Public setup documentation and cross-linked Jira configuration pages are now available. *** ### Upgrade Notes * **Azure DevOps:** To use the new Azure DevOps integration, create an ADO connection under Config → Connections using a Personal Access Token with Boards, Code, and Build scopes. All three surfaces (Boards, Repos, Pipelines) are available through a single connection per organization. * **Microsoft Teams:** Teams messaging requires a Microsoft Graph OAuth connection. Webhook-based notifications require only a webhook URL and work without an OAuth flow. Refer to the Microsoft Teams Webhook setup guide in the documentation. * **PAT connection permissions:** Roles that held the previous coarse-grained PAT connection permission are automatically migrated to all five category-scoped permissions. Review role configurations post-upgrade if you want to restrict specific categories. * **Custom CA certificates (Helm):** The `customCACerts` feature requires Helm chart v1.19.42 or later. Refer to the AWS deployment guide for configuration details. * **Cortex REST embedding (SPCS):** No action required. The SQL-based embedding fallback is now applied automatically on 403 responses. * **Validate after upgrade:** Confirm that thread compression completes successfully on at least one long thread, that chat streaming shows no tool-envelope artifacts, and that any Azure DevOps or Microsoft Teams connections authenticate correctly.