Genesis v1.19.96 Release Notes
October 3, 2026Highlights
- Costs now tracks spending on every installation, showing when recording began and separating estimated-price spend from catalog-priced spend.
- Context Graph now shows which pipelines write a schema’s tables, including stored procedures and Talend jobs.
- The dbt tool can now convert a Snowflake or SQL Server view into a dbt model automatically, flagging any it can’t verify.
- OpenAI defaults use GPT-6.1 Sol/GPT-6 Luna; GPT-6 Astra is selectable. Azure defaults stay unchanged; retired models are removed and unpinned, with explicit rejection errors.
- Text sent during a turn now steers the agent at its next tool result without interrupting work; attachments still wait until the turn ends.
- Requests now lists mission questions alongside approvals; answer directly in the drawer to continue a mission, or dismiss and restore items.
Deployment Artifacts
Use these exact references when deploying this release:- Genesis release:
v1.19.96(promoted fromv1.19.96-rc.21) - Docker Hub image:
genesiscomputingai/genesis:v1.19.96 - Docker Hub digest:
genesiscomputingai/genesis@sha256:8a0898d5b3dbdb312c796ca23e60bd14a5c7cea2e55843d364f67c398f04e500 - Helm chart:
oci://registry-1.docker.io/genesiscomputingai/genesis - Helm chart version:
0.3.108(appVersion: v1.19.96)
Upgrade Notes
- Cluster, Terraform, Azure, Databricks, and local Kubernetes deployments on v1.19.95 should upgrade to restore configured AI providers.
- Deployments using retired o1, o1-mini, or o1-preview models should select a supported model before upgrading.
- The Secrets panel is visible by default; set features.show_secrets_panel to false if administrators need it hidden.
Improvements
- Context Graph now scans Informatica PowerCenter exports to trace tables back to the workflows and sessions that produce them.
- The dbt tool can now plan multi-view migrations and generate dbt source and schema YAML files automatically.
- Leftover upgrade snapshots are now deleted automatically after seven days when retention is off, while snapshots that may be the only safe copy are preserved.
- Connections routed through an Outpost no longer show a false Snowflake egress requirement or a disabled test button.
- Snowflake Native App installs now show install stage, progress percentage, and estimated time remaining, so a slow cold start no longer looks broken.
- Work and Settings now show their screens in rail popovers instead of a second navigation rail, with the Missions icon visible in dark menus.
- Amazon MWAA connections now read DAGs, failures, and logs through AWS IAM, including private webservers; Native App end-to-end validation remains pending.
- Snowflake Native Apps now expose Genesis to consumer Cortex Agents through an automatically registered MCP server, where the account and region support it.
- MCP Gateway now includes Cortex Code setup instructions tailored to each deployment’s authentication method.
- The dbt tool can generate models in your chosen SQL house style, then re-verify each styled model against the original query.
Bug Fixes
- Usage tracking now retries cost record writes when the local cost store is busy, instead of dropping them immediately.
- Inactive AI providers now keep their own embedding model setting instead of overwriting the active embedding configuration.
- Cost summaries, Agent Review, and budget views no longer show unpriced usage, like image or speech generation, as a false zero cost.
- Chat sidebar grouping and search by person now list every thread you can see, not just recent ones.
- A fresh Postgres metadata database now migrates successfully to the latest schema, and concurrent startups wait for each other instead of failing.
Preview
- Administrators can enable Genesis Outposts to connect agents to systems inside a customer’s private network through an outbound-only connection.
- Snowflake PAT and key-pair connections can optionally log in over Snowflake’s internal network, enabling scheduled jobs on PrivateLink accounts without internet access.
- Native App administrators can call CORE.HIBERNATE_INSTANCE to suspend services and compute pools until explicitly started; full hibernate-to-start validation remains pending.

