> ## 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.

# v1.19.83

> Genesis Data Agents v1.19.83 release notes

## Genesis v1.19.83 Release Notes

**July 18, 2026**

Data comparison and migration workflows, stronger Context Graph lineage, LiteLLM-backed Azure and Databricks providers, and broad reliability improvements across chat threads, blueprints, connections, and platform identity.

#### Data Comparison and Migrations

Genesis now supports structured data comparison and end-to-end migration tracking so teams can validate changes before and after cutover.

* A new **data comparison** framework lets agents compare query results across connections with dialect-aware adapters, reconciliation tracking, watermark-based snapshots for moving data, and scheduled re-checks that surface drift over time.
* Comparison workflows integrate with mission blueprints so validation can run as part of automated migration playbooks.
* The **Launchpad migration wizard** collects source and target connections up front so missions start with real work instead of failing on missing context; plan-only runs end with clear execute-later instructions.
* **Migration tracking** connects planning, execution, and validation in one workflow so users can follow each migration from start to finish and confirm results at every stage.
* Snowflake is supported as a second comparison dialect, validating cross-engine comparisons beyond the initial pair.

#### Context Graph

Scanning and lineage in the Context Graph are more complete and accurate across Git, SharePoint, and dbt projects.

* Git-based scan sources now report accurate scan state instead of always showing empty status.
* **dbt lineage** discovered in Git repositories is automatically wired onto scanned database nodes, so transformation dependencies appear in the graph without manual linking.
* Scanner lineage summaries no longer fail on malformed prompts or argument mismatches during estate generation.
* **SharePoint** scans on same-tenant multi-site deployments no longer collide source identifiers, so each site is indexed independently.

#### AI Providers and LLM

Azure OpenAI and Databricks models route through the LiteLLM engine for more consistent behavior across deployments, with targeted fixes for model assignment and structured output.

* **Azure OpenAI** now defaults to the LiteLLM engine, with all remaining Azure paths routed through the same stack for consistent tool calling and streaming.
* **Databricks** models route through LiteLLM, including recovery of structured `response_format` content that was previously lost in synthetic tool-call translation.
* Databricks LLM configuration is unblocked on Snowflake SPCS deployments, with approve-external-access guidance and support for secret-only custom endpoint providers.
* OpenAI tier models resolve from live model-assignment settings instead of stale hardcoded defaults.
* Operator model-assignment overrides are preserved when re-saving LiteLLM proxy configuration.
* Reasoning-effort parameters are no longer sent to Databricks and other non-reasoning models, preventing spurious API errors.
* Snowflake Cortex context-window limits are restored in the conversation trimmer so long threads compress correctly.
* AI Providers UI polish: dropdown clipping, dark-mode controls, embedding edit menus, chat-override save behavior, and embedding model list handling are corrected.

#### Chat, Threads, and Agents

Thread continuity, navigation, and agent identity are hardened so bookmarks, handoffs, and multi-bot workflows behave predictably.

* **Fork a thread** from any message and **edit-and-resume** prior messages to explore alternatives without losing the original conversation.
* Opening a thread or bot URL that no longer exists redirects to the agents list instead of showing a blank or broken page.
* Thread actions — pin, rename, delete, and fork — work on legacy chat links that omit the agent from the URL.
* Bots are addressed by canonical **bot ID** everywhere; display names and wrong-case IDs now return a clear not-found instead of routing to the wrong agent.
* One-on-one check-ins no longer open in the wrong bot's or wrong user's thread.
* The chat **stop** button responds during stalled LLM streams instead of waiting indefinitely for the next token.
* **Find in thread** search matches text across markdown boundaries so formatted messages are fully searchable.
* Virtuoso's message list self-heals when a thread loads blank due to stuck hidden messages.
* Configured bot tools are preserved across upgrades and registry refreshes instead of being dropped.

#### Blueprints and RBAC

Blueprint administration and role-scoped access are expanded so managers can govern workflows without full user-admin rights.

* New **view all blueprints** and **manage all blueprints** privileges let administrators scope blueprint visibility and control separately from other admin roles.
* Blueprint managers can grant role access to their blueprints without requiring full user-admin permission.
* Blueprint kickoff tool groups respect the bot allowlist so restricted agents can start approved workflows.
* Users with **view connections** permission can see the Data tab and connection list instead of an empty panel.
* When authenticated through an OAuth proxy, Genesis uses the user's web-login account and refuses unknown emails instead of creating orphan accounts.
* Dashboard health probes are exempt from OAuth-proxy authentication, with per-request CSRF protection enabled for browser sessions.

#### Connections and Integrations

Connection setup, AWS identity, and data-browser navigation are clearer and more reliable.

* **AWS IRSA** connections show the IAM role ARN in the edit UI and during new-connection selection, with dark-mode styling and caller IAM ARN included in access-denied errors for faster troubleshooting.
* The default AWS Insight connection uses IRSA authentication when deployed with a pod service account.
* **Databricks deep links** from the Context Graph resolve correctly in the Data Browser, including legacy links with mismatched connection IDs and case-insensitive identifier matching.
* Connection scope changes on update are no longer silently dropped; Confluence access-vs-credentials errors and stale support-bundle regeneration are also corrected.
* The legacy system Snowflake SPCS connection is hidden from agents and dashboards.
* **JAMS** connections support HTTP Basic authentication for IIS/Windows-gated servers.
* Large **GitLab** file reads are no longer silently truncated, preventing partial commits; missions can also commit large files to a GitLab project by ID when the mission repo has no remote configured.

#### Missions and Journeys

Mission workflows surface more context and history so teams can trace work and discover repeatable patterns.

* The Missions list adds **Load more** so older missions remain reachable beyond the initial page.
* **Journey phases** show database activity grouped by phase so users can trace which queries ran at each step.
* Genesis captures privacy-safe **workflow history** to recommend repeatable journeys based on how teams actually work.
* Exporting missions and agents to ZIP archives writes files correctly instead of producing corrupt packages.

#### Knowledge Base and Document Index

* Document indexes support configurable **chunk size and strategy**, with a **RECHUNK** action to rebuild existing indexes under new settings.
* Wiki LLM prompts are capped to prevent context-window overflow during learning workflows.

#### Skills

* Uploaded skill ZIP packages ignore common editor and operating-system artifacts (such as `.DS_Store` and `__MACOSX`) instead of rejecting the entire package.

#### Thread Import

* Imported thread messages record the verified importing user as the author so per-user counts and attribution stay accurate.

#### Reliability and Platform

* Blueprint registration writes are atomic, preventing managed sources from being wiped during transient storage errors.
* A thread-directory reconciler quarantines stray and case-variant bot folders to keep thread storage consistent.
* Thread metadata is case-sensitive and driven strictly by the database, preloaded at startup for predictable lookups.

#### Security

* Container images include dependency updates addressing known vulnerabilities in bundled components.

### Deployment Artifacts

Use these exact references when deploying this release:

* **Genesis release:** `v1.19.83` (promoted from `v1.19.83-rc.6`)
* **Docker Hub image:** `genesiscomputingai/genesis:v1.19.83`
* **Docker Hub digest:** `genesiscomputingai/genesis@sha256:f3bfbbb68f998d1606c979a2ea66a0345331afbde78fed9eb4f9204acc63a3d4`
* **Helm chart:** `oci://registry-1.docker.io/genesiscomputingai/genesis`
* **Helm chart version:** `0.3.59` (`appVersion: v1.19.83`)

### Upgrade Notes

* **Data comparison:** Run a comparison between a source and target connection through a mission blueprint or agent workflow. Confirm results appear in the reconciliation view and that a scheduled re-check surfaces any drift.
* **Migrations:** Open the Launchpad migration wizard, select source and target connections, and run a plan-only migration. Confirm it ends with execute-later instructions. Run a full migration and verify planning, execution, and validation stages are tracked end to end.
* **Context Graph:** Run a Git scan that includes dbt models and confirm lineage edges appear on database nodes. Re-run a SharePoint multi-site scan and verify each site is indexed separately.
* **AI Providers:** On Azure and Databricks deployments, confirm chat and structured-output calls succeed through the LiteLLM engine. Re-save LiteLLM proxy settings and verify model assignments persist. On SPCS, configure a Databricks provider and confirm external-access approval links appear.
* **Chat threads:** Fork a thread from a mid-conversation message and edit-and-resume a prior turn. Open a bookmarked thread URL without an agent segment and confirm pin/rename/delete still work. Open a deleted thread URL and confirm redirect to the agents list.
* **Blueprints:** As a blueprint manager without user-admin, grant role access to a blueprint and confirm it succeeds. Verify users with view-connections see the Data tab.
* **Connections:** Create or edit an AWS IRSA connection and confirm the IAM role ARN displays. Click a Databricks deep link from the Context Graph and confirm the Data Browser opens the correct object.
* **Missions:** Scroll the Missions list and use Load more to reach older entries. Open a journey-phase view and confirm database activity is grouped by phase.
* **Skills:** Upload a skill ZIP that includes `.DS_Store` or similar artifacts and confirm the package is accepted.
* **Validate after upgrade:** Send a test chat message, run a short Context Graph scan, and confirm AI provider status and model assignments on the providers overview page.

***

**— The Genesis Team**