Genesis v1.19.81 Release Notes
July 11, 2026 A redesigned AI Providers experience, stronger Context Graph scanning and troubleshooting tools, encrypted connection migration, and broad reliability fixes across embeddings, thread import, RBAC, and platform integrations.AI Providers
The AI Providers settings page is rebuilt around a guided setup wizard and a clearer at-a-glance overview of each provider’s status.- A new AI Providers configuration page walks administrators through provider setup in guided steps — credentials, connection test, model assignment, and optional embeddings — with per-provider forms for all supported backends.
- A single overview shows each provider’s setup stage (active, needs models, ready, and so on) so you can see what still needs attention without opening every connection.
- Deleting a provider now fully removes its stored credentials and clears stale active pointers, so removed providers no longer reappear as configured after restart.
- LiteLLM-family providers (Anthropic, Gemini, Bedrock) activate reliably instead of failing silently when aliases are used.
- Switching or removing a chat provider no longer resurrects OpenAI as the embedding provider on the next restart.
- On Snowflake Cortex and SPCS deployments, embedding configuration no longer silently falls back to OpenAI when the saved provider and model pair are incompatible — invalid combinations are corrected and validated on save and at startup.
- Azure OpenAI connections can be saved before a deployment name is chosen, matching the wizard’s two-step flow.
- UI polish: dropdown clipping, dark-mode controls, embedding edit menus, and chat-override save behavior are corrected.
Context Graph
Scanning and troubleshooting the Context Graph is more transparent and easier to validate before you run a job.- A Check permissions action on each connection in the scan configuration modal validates required database and API privileges before scheduling a scan, with per-check pass, warn, or fail results and remediation guidance for Snowflake, Databricks, GitHub, GitLab, Postgres, and SQL Server.
- Administrators with config-panel access can open a live log viewer from the scan status modal to search, filter, and download Context Graph engine logs without server access.
- Personal-scope connections are visibly greyed out in the scan connection picker with a clear explanation that shared service-account connections are required.
- Git-based scan sources now report accurate scan state instead of always showing empty status.
- Context Graph log downloads work on authenticated deployments.
- Context Graph structured-output and wiki learning workflows authenticate correctly through OAuth and static-header LiteLLM proxy gateways.
Connections and Integrations
- Administrators can export and import encrypted connection bundles to migrate selected connections — including resolved secrets — between Genesis instances using a passphrase-protected file with a 24-hour expiry.
- dbt Cloud connections now support Personal Access Token (Bearer) authentication in addition to service tokens.
- Connection setup screens use plain-language scope controls and grouped access options so it is clearer what each permission grants.
- Power BI connections auto-detect the admin API for tenant-wide workspace discovery.
- Databricks Genie can be connected as an MCP server over HTTP, letting agents query Genie spaces from chat.
- Snowflake file uploads route to the correct database for the selected connection instead of silently landing in the wrong place while reporting success.
- The legacy system Snowflake SPCS connection is hidden from agents and dashboards so it is no longer suggested as a usable connection.
RBAC and Identity
- Role assignments now support Azure AD App Roles alongside security-group GUIDs, so enterprises using app-role-based access control can map roles without workarounds.
- The profile dropdown shows an OIDC token claims popover so administrators and support staff can inspect the identity attributes Genesis received at login.
- Access-preview results in the permissions UI now match what the API returns.
Missions
- The Missions list adds a Creator column with sorting and filtering so you can find missions by who created them.
Chat and Agents
- Tool-call display in chat is simplified and grouped more clearly using server-assigned turn identifiers.
- Agents have a dedicated tool for reading Knowledge Base documents, reducing confusion with the general Genesis documentation reader.
- Configured bot tools are preserved across upgrades and registry refreshes instead of being dropped.
Thread Import
- Coding-agent conversation threads in JSONL format can be imported into Genesis for continuity across tools.
- Imported threads record the verified importing user as the message author, so per-user counts and attribution stay accurate.
- Thread import identity is verified through the OAuth proxy so the claimed author cannot be forged.
Wiki and Memory
- Wiki memory surfaces top learning headlines with improved ranker telemetry and scheduled health reporting so operators can confirm the learning pipeline is active.
Configuration and Reliability
- Replacing configuration sections no longer silently resets the active LLM provider when a partial write fails mid-operation.
- External secret references (AWS Secrets Manager, Vault, and similar) resolve correctly in the LLM configuration pipeline.
- Thread compression on Databricks-backed deployments handles authentication failures gracefully with a safe local fallback instead of failing outright.
- Managed source sync hydrates skill content from a local object store pinned to the release commit, reducing per-artifact remote fetches and speeding large syncs.
- Auto-send of support bundles now requires explicit operator consent and is off by default; the receiver URL is pre-filled when enabled.
- The public health endpoint no longer exposes personally identifiable information; detailed slot diagnostics are available through an authenticated admin endpoint.
Security
- Container images include dependency updates addressing known vulnerabilities in bundled components.
Upgrade Notes
- AI Providers: Open the new AI Providers page and confirm each provider shows the correct setup stage. Walk through the wizard for any provider still marked not configured. Delete a test provider and restart to confirm it does not reappear. On Cortex/SPCS, verify embedding settings stay on your chosen provider after save and restart.
- Context Graph: Open scan configuration, run Check permissions on a Snowflake or Databricks connection, and confirm results before scheduling. Open the log viewer from an active scan and verify lines stream and download works. Confirm personal connections appear greyed out with the service-account message.
- Connection bundles: As an admin, export a small set of test connections with a passphrase, preview the bundle on a staging instance, and apply with conflict resolution. Confirm secrets resolve and connections work after import.
- dbt Cloud: Create or edit a connection using a Personal Access Token and confirm test connection succeeds.
- Databricks Genie: Add a Genie MCP connection over HTTP and send a chat message that queries a Genie space.
- Snowflake uploads: Upload a file through a non-default Snowflake connection and confirm it lands in the expected database.
- RBAC: If using Azure AD App Roles, map a role and confirm the user receives the expected permissions. Open the profile dropdown and inspect OIDC claims after login.
- Missions: Sort and filter the Missions list by Creator and confirm results match expectations.
- Thread import: Import a JSONL thread and confirm message authorship and per-user counts reflect the importing user.
- Config: Apply a partial config replacement and confirm the active LLM provider is unchanged. Verify external secret references resolve for LLM providers that use AWS or Vault refs.
- Validate after upgrade: Run a Context Graph permission check and a short scan, send a test chat message on a proxy-backed thread, and confirm AI provider status on the new overview page.
— The Genesis Team