> ## 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.

# v1.19.95

> Genesis Data Agents v1.19.95 release notes

## Genesis v1.19.95 Release Notes

**September 25, 2026**

### Highlights

* Coding tasks can now run on a new Sandbox Runner backend, in a dedicated Docker container or Kubernetes Pod, once enabled in Advanced Settings.
* Administrators can now snapshot an instance to S3 or Google Cloud Storage and restore it onto Kubernetes, including moving a Snowflake SPCS instance.
* Action Review is now the single approvals destination, on by default, and the Requests page lists approvals waiting on you first so you can apply them in place.
* Databricks agents can now list Unity Catalog metric views, and changes to a metric view are risk-assessed instead of refused.
* Flows can now bulk-copy a whole table into Databricks or SQLite, with substituted values kept safely inside their literal.
* A new read-only DataStage 11.7 connector discovers projects and jobs and takes in original DSX exports.
* New threads, active threads, and mission kickoff now share a persistent working-context control for choosing a repository or folder, connections, and a model tier.
* MCP clients can now hand a task to a Genesis agent with the ask\_agent tool, optionally waiting for its answer or stopping it.

### Deployment Artifacts

Use these exact references when deploying this release:

* **Genesis release:** `v1.19.95` (promoted from `v1.19.95-rc.18`)
* **Docker Hub image:** `genesiscomputingai/genesis:v1.19.95`
* **Docker Hub digest:** `genesiscomputingai/genesis@sha256:27677f6e4e4399afeadbbfa64335eaa94b0e64574bbae6f36cf053ce1279cee5`
* **Helm chart:** `oci://registry-1.docker.io/genesiscomputingai/genesis`
* **Helm chart version:** `0.3.102` (`appVersion: v1.19.95`)

### Upgrade Notes

* If you never chose a coding backend, coding tasks now use the Sandbox Runner, which is off by default; enable it or select claude\_code in Advanced Settings.

### Improvements

* Action Review now governs more writes, including source-control, warehouse, and Databricks grant changes, and opens each approval with a plain decision brief.
* Data Estate Health dashboards now lead with visual comparisons, supported priorities, and affected consumers, keeping detailed evidence in disclosures.
* Flow bulk-copy steps can now load directly into Snowflake through a native staged-file loader versus a row-by-row write.
* A failed write to an external target now shows the target's real error and offending column in the Flows UI, with row values masked.
* Chat now pins the message that started a turn while you read its answer, and browser tabs show the active thread or mission name.
* MCP sign-in now needs one registered callback URL per deployment, and the MCP Gateway page's Claude Code setup includes an explicit sign-in step.
* Foundry migrations now carry over ingestion settings and schedules, and Talend job descriptions return mapping and SQL rule evidence with credentials redacted.
* Cost tracking now includes image generation, speech, web search, Cortex, and document parsing, and Pulse Costs shows a waiting message until the first cost summary is ready.

### Bug Fixes

* Chat requests that hit provider rate limits now retry automatically within a configurable window, 120 seconds by default, honoring the provider's suggested delay.
* When Snowflake refuses a Cortex model under Model RBAC, Genesis now shows Snowflake's own remediation guidance instead of a generic error.
* The Advanced Settings page now opens in seconds instead of minutes on Snowflake-hosted deployments.
* Restricted Caller's Rights queries now ask users to sign in again when their Snowflake session expires, instead of showing a generic connection error.
* Sign-in no longer fails over a trailing-slash issuer mismatch, and MCP sign-in accepts Claude, Cursor, and VS Code callbacks without configuration.
* Concurrent bulk copies no longer freeze the server, and the backend now exits within a bounded deadline after a stop signal.
* Messages delivered to an idle thread, such as a finished flow or handoff report, now appear within about a second instead of up to 30 seconds.
* Flow copies into Databricks no longer strip apostrophes, and bulk copy keeps numeric columns numeric by loading empty strings as NULL.