Skip to main content

​
Genesis v1.19.55 Release Notes

May 28, 2026 Wiki commentary, a unified Context Graph and wiki view, post-upgrade guidance, managed-artifact administration, and broad reliability improvements across chat, compression, MCP, and Snowflake deployments.

​
Context Graph and Wiki

The wiki and Context Graph experience gained a side-by-side layout and a new way for teams to annotate knowledge directly on wiki pages.
  • The Digital Twin view now shows the Context Graph and its wiki pane together in a resizable split layout. Clicking a graph node opens the matching wiki page; clicking a wiki link pans the graph to that asset.
  • Every wiki page — entity pages and subgraph documents — now supports a Commentary section where users with read access can post markdown notes. Authors can remove their own comments; removed comments leave a visible tombstone.
  • Legacy operator-authored learnings submission is retired in favor of page-level commentary; existing learnings content remains readable.
  • Six wiki feature toggles moved from environment variables into AppConfig, so operators can turn wiki writing and related behaviors on or off from Advanced Settings without a server restart.

​
Post-Upgrade Guidance

Administrators see clearer guidance immediately after upgrading.
  • After each upgrade, Genesis can generate a personalized “what’s new” summary tailored to the install profile and surface it in the upgrade notice modal alongside static release notes.
  • Admins can also request an on-demand summary from the same modal at any time.

​
Genesis Apps, Skills, and Flows: Managed Sources

Repository-backed artifact management is easier to administer and safer when one Git repo backs multiple artifact types.
  • A unified Managed Sources admin page lists every git-backed source and which artifact types each repo serves, fixing cases where registering the same repository through separate panels silently overwrote earlier registrations.
  • Operators can fork an App from the dashboard and push unmanaged Apps into a managed repository through guided UI flows — no browser prompt dialogs.
  • A new admin permission controls who can view and edit managed-source configuration.

​
Chat, Threads, and Compression

Chat responsiveness and thread reliability improved across polling, streaming, sharing, and long-thread compression.
  • The chat thread view no longer re-renders on every background activity poll when nothing visible changed, eliminating idle UI jank on busy deployments.
  • Stuck “Joining thread stream…” banners clear correctly when a subscription retry is abandoned.
  • Thread share metadata handles missing owner and visibility fields without returning errors.
  • Long-thread compression now respects the AI-panel model selection for non-OpenAI providers (Cortex, LiteLLM Bedrock, and others) instead of silently falling back to defaults in worker processes.
  • Compression history storage uses a compact schema with unlimited entries, and the compression history panel shows monotonically numbered events including manual compressions.
  • Auto-generated thread titles recover correctly when metadata flush workers restart.

​
Hosted MCP and External Clients

Hosted MCP and third-party client integrations received OAuth, discovery, and visibility fixes.
  • Hosted MCP supports an optional OAuth authorization-server proxy for clients that do not follow delegated-discovery standards.
  • Per-thread MCP resource catalogs are bounded and durable, stabilizing discovery for long conversations.
  • Direct MCP tool calls render immediately in the chat UI, and REST-submitted prompts appear without delay.
  • A “Show MCP” filter in the thread sidebar helps operators focus on MCP-origin threads.
  • Azure OIDC user resolution and OAuth metadata alignment fix authentication failures for enterprise identity providers.

​
Connections, LLM Providers, and Cortex

Connection and provider configuration gained new auth modes and admin controls.
  • LiteLLM proxy connections support a network-identity authentication mode for gateways reached through Tailscale, service mesh, or reverse-proxy trust — no API key or Authorization header is injected.
  • An internal OAuth wrapper preserves service-account header state when LiteLLM proxy connections go stale, preventing silent auth failures.
  • Per-model LLM pricing overrides can be set in AppConfig so cost estimates use operator-defined rates instead of catalog defaults alone.
  • Cortex Code connections support Snowflake key-pair authentication, and CLI errors from Cortex now surface in chat instead of empty responses.
  • Cortex Test Connection and Save work correctly on SPCS-native deployments where empty-string fields were previously rejected.
  • PDF previews in the dashboard render correctly — the content security policy now allows blob URLs in embedded frames.
  • File exports with non-ASCII filenames download with correct names instead of garbled headers.

​
Missions and Scheduler

Mission Center and scheduled jobs gained quality-of-life improvements.
  • A new “All Not Completed” filter shows every mission still in play — pending, running, stuck, or paused — in one view.
  • Mission tag filtering uses OR semantics: selecting multiple tags shows missions matching any selected tag, and each tag chip can be removed individually.
  • Scheduled job names are editable inline from the job detail modal and the scheduler list.

​
RBAC and Administration

  • The admin audit log is now browsable, filterable, and exportable from the dashboard, completing the read side of governance auditing introduced in the prior release.
  • Connections page filtering controls were refined for faster discovery across large connection inventories.

​
Documentation Access

  • Documentation links opened from within Genesis authenticate automatically — users land on the docs site signed in without entering a separate password.

​
Snowflake Native App and Private Networking

Snowflake deployments are safer to upgrade and can reach private LLM endpoints.
  • Snowflake native app upgrades no longer silently drop and recreate existing services when the deployed spec version is stale; incompatible upgrades abort and keep the prior version running, protecting block-storage volumes.
  • Optional embedded Tailscale userspace networking lets SPCS instances reach private LLM gateways over WireGuard without kernel privileges or spec-version bumps — existing customers see no change until the feature is enabled.

​
Helm and Kubernetes (Self-Hosted)

Self-hosted Kubernetes deployments gained operational hardening.
  • Pods drain gracefully on shutdown with improved SIGTERM handling and configurable termination grace periods.
  • Extension slots allow sidecar and init containers without forking the Helm chart.
  • LLM credential values referenced from AWS Secrets Manager or GCP Secret Manager resolve automatically at runtime.