Skip to main content

​
Genesis v1.19.87 Release Notes

August 14, 2026 Highlights:
  • Claude Opus 5 and Sonnet 5 are available across Cortex, Anthropic direct, and Amazon Bedrock. If you use Genesis defaults for your model and have selected one of these providers, you will be set to Opus 5 or Sonnet 5 by default. To revert to a different model, change it on the model configuration page. The LiteLLM Proxy remains unchanged, and you can configure Opus 5 or Sonnet 5 there manually if needed.
  • A full Azure DevOps integration lets agents read and update Boards, Repos, and Wiki content, with work items flowing into Requests.
  • Migration Verification gains SQL Server support, cross-engine comparisons, a redesigned Human Review experience, and an adjudication workflow for disputed findings.
  • Context Graph scanning now has on-by-default spend caps and cost visibility in the usage dashboard.
  • Jira and Requests are more reliable, with automatic intake from Slack and Microsoft Teams.
This release also improves AI provider support, agent file tools, configuration management, and platform security.

​
Azure DevOps

Agents can now work with Azure DevOps across Boards, Repos, and Wiki from a single connection suite.
  • Agents can search, read, and navigate Azure DevOps Wiki pages, including revisions, attachments, and the page tree for both project and code wikis.
  • Agents can query Azure Boards work items, search with natural language that translates to WIQL, and read sprints, comments, history, and links.
  • Agents can search Azure Repos, browse branches and commits, read pull requests, and access Advanced Security findings.
  • Agents can create and update Wiki pages, open branches, add pull request review comments, record review votes, link work items, and move items between sprints.
  • Azure Boards work items and assignments appear on the Requests page as agent-managed tracked work, so operators can see DevOps assignments alongside other intake channels.
  • Assignment threads keep the correct automation owner when an agent reopens existing work, and automation context is preserved across updates.

​
Migration Verification

Migration Verification received a major upgrade to how comparisons run, how findings are tracked, and how operators review results.
  • SQL Server is now a first-class platform for estate characterization, comparison, and profiling, with T-SQL-aware column templates.
  • Cross-engine comparison pairs are available out of the box for SQL Server to Snowflake and SQL Server to Databricks migrations, including platform attribution and collation-aware distinct counts.
  • Human Review is now the primary entry point for reviewing comparison results, with a full-viewport visualization modal, live run watching, and persistent attributed decisions.
  • Findings are tracked in a durable ledger across runs, with plain-language labels, behavioral coverage, cause grouping, and an Issues view annotated by each run.
  • An adjudication workflow lets operators dispute findings, with implementer explanations, independent verification, receipts, and conversions visible in Human Review.
  • A ruling library stores estate-specific decisions that agents and reviewers can reference when the same question arises again.
  • The estate page is now a dossier that composes comparison status, ledger findings, and labeled objects in one place.
  • The Prover landing page is reorganized into Comparisons and Sources sub-tabs, with tile and list views, inline actions, and faster estate list loading.
  • Characterization scans resume after interruption instead of restarting from zero, and scan progress is preserved across server restarts.
  • Long-running scans and comparisons stay alive through heartbeat improvements, and Databricks scans run faster with fewer stalls.
  • Verdicts abstain when evidence is insufficient, and Genesis refuses to approve a comparison when unresolved findings are hidden by a truncated worklist.
  • Estates can be archived durably with a server-backed flag instead of browser-only state.
  • In-product spec authoring lets operators compile, seal, and review migration specifications with provenance tracking and a review interface.

​
Context Graph

Operators gain more control over scanning cost and resilience.
  • Context Graph scanning is bounded by default with an on-by-default spend cap and a configurable scanner tier.
  • Scan spend appears in the cost and usage dashboard, so operators can see Context Graph costs alongside other platform usage.
  • Custom scanner and scheme lifecycle management lets operators register, update, and retire custom metadata schemes from the app.
  • Document scanners are more resilient, with improved transport handling, per-item fault tolerance, and bulk fetch support.
  • Context Graph logs persist on the deployment volume with bounded multi-day rotation for easier troubleshooting.
  • Scan progress no longer resets to zero after a restart when work was already in flight.

​
Requests and Work Tracking

Requests and connected work-tracking systems are easier to configure and more reliable in daily use.
  • The Requests page has a new name, icon, and configure path, and can be hidden under UI Complexity settings for simpler deployments.
  • Slack and Microsoft Teams default to automatic intake for new Requests configurations, so messages flow into tracked work without extra setup.
  • Jira polling is more resilient to schedule drift and eventual consistency, and the enhanced search contract is corrected for Jira Cloud.
  • Jira Request titles refresh on duplicate polls instead of staying stale, and Genesis fails closed when a Request lookup returns empty.

​
AI Providers and Documents

Provider configuration, model availability, and document indexing are more predictable.
  • Claude Opus 5 and Sonnet 5 are available across Cortex, Anthropic direct, and Amazon Bedrock. Deployments on Genesis model defaults with one of these providers move to Opus 5 and Sonnet 5 automatically; the model configuration page reverts that choice. The LiteLLM Proxy is unaffected and can be pointed at either model manually.
  • Bedrock credential fallback now passes session tokens when operators authenticate with temporary AWS credentials.
  • The embeddings UI no longer claims no model is configured while indexes are still loading or when loading fails; the active embedding model is listed once loading completes.
  • Embeddings no longer fall back to an OpenAI model nobody configured, and stranded document corpora can be re-indexed.
  • LiteLLM Proxy can be activated after setup without a missing-configuration error.
  • Document indexes can be deleted from every collection that holds them, not just one collection at a time.

​
Agents, Tools, and Git

Agents work with files, repositories, and threads more reliably.
  • Agents can read a bounded window of a file, including from the end of the file, and always receive content instead of an opaque spill reference.
  • Thread export and import now carry query results and V6 compression artifacts, so archived tool payloads survive a move between environments.
  • Agents can open a thread found by search even when it belongs to another bot, subject to the same permission checks as before.
  • Skills can be deleted, and registry-only skills show a description fallback instead of appearing as permanent broken stubs.
  • Agent skills can be validated locally to catch broken metadata, links, and triggers before deployment.
  • Concurrent Git file writes no longer fail when multiple agents touch the same repository.
  • Managed sources can exclude skills, Genesis Apps, and flows from checkout; an excluded flow no longer runs on its schedule.
  • Untracked worktree state is preserved when agents work in Git worktrees.

​
Connections and Configuration

Configuration changes are safer, faster, and easier to understand.
  • Opening the Config Profiles page no longer freezes the server on large configurations.
  • Imported connections keep the secret’s original owner, not the person who performed the import.
  • Imported blueprint copies are immediately accessible because local ownership is assigned on import.
  • The Permissions panel shows plain-language descriptions and hover tooltips for every permission.
  • Each user’s sort order is remembered for Knowledge Bases, Blueprints, and Skills lists.
  • Connection row actions stay visible on hover instead of shifting as the pointer moves.
  • AWS AssumeRole works from pod IRSA for deployments that use Kubernetes service account credentials.
  • SQLite write queries commit immediately, so agents no longer strand the database write lock.
  • Teams external-access approval targets the correct reference instead of approving the wrong object and reporting success.
  • Bot role update errors are surfaced clearly instead of failing silently.

​
Chat, Threads, and Microsoft Teams

Chat navigation and Teams presentation are improved.
  • The chat sidebar splits agent and thread lists so threads stay visible when many agents are configured.
  • Agent and thread actions are one click away in the sidebar, with always-visible controls on touch devices.
  • WebSocket degradation signals are accurate and calmer, so users see realistic connection status during network issues.
  • Teams bots show their initial as an avatar instead of a blank square.
  • The Teams SPCS relay ships as a standalone deployable service for Snowflake-hosted Teams integrations.

​
Flows, Missions, and Genesis Apps

  • Scheduled flows that recover and then fail again are reported, instead of going silent after the first recovery.
  • Repeated flow-failure alerts are deduplicated, so operators are not notified once per failed run for the same issue.
  • Presenter coding-agent choices respect availability, so demo workflows only offer agents that can actually run.
  • dbt recovery in presenter demos is durable across session restarts.

​
Matillion Connector

  • Agents can list Matillion pipeline executions and pipelines for run-history and discovery.
  • Agents can list Matillion connections, with automatic fallback when only one connection exists.

​
Platform Operations

  • Managed sources pinned to a missing release branch fail the sync instead of silently falling back to the default branch.
  • Excluding a file or folder from a managed source checkout removes it from the Git page view.
  • The dashboard version selection survives a page refresh.
  • Output-less tool calls no longer render as red failures in the chat transcript.
  • CRM merge probes no longer rebuild a Snowflake engine every minute, reducing unnecessary load.

​
Security

  • Container images include updates to GitPython and other dependencies addressing known high-severity vulnerabilities.
  • Operating-system package updates address additional known vulnerabilities in bundled components.

​
Deployment Artifacts

Use these exact references when deploying this release:
  • Genesis release: v1.19.87 (promoted from v1.19.87-rc.6)
  • Docker Hub image: genesiscomputingai/genesis:v1.19.87
  • Docker Hub digest: genesiscomputingai/genesis@sha256:df23a5fea204037a5794b5579713918d4a5fc3fb193d5aec1207fdb0544cad6c
  • Helm chart: oci://registry-1.docker.io/genesiscomputingai/genesis
  • Helm chart version: 0.3.66 (appVersion: v1.19.87)

​
Upgrade Notes

  • Azure DevOps: Configure a connection with Boards, Repos, and Wiki access. Ask an agent to search a work item, read a Wiki page, and create a test comment on a pull request. Confirm assignments appear on the Requests page.
  • Migration Verification: Open an existing estate and confirm the dossier view shows comparisons and findings together. Run a comparison and walk through Human Review, including the visualization modal. If you use SQL Server, register an estate and confirm characterization completes. Archive a test estate and confirm the archived state persists after a browser refresh.
  • Context Graph: Review the cost and usage dashboard for scan spend. Confirm the default spend cap is appropriate for your environment before the next scheduled scan.
  • Requests: Verify Jira and Azure DevOps work items sync to Requests with current titles. If you use Slack or Teams intake, send a test message and confirm it creates or updates a Request automatically.
  • AI Providers: Send a test chat through Bedrock with temporary AWS credentials. Open the embeddings settings and confirm the active model is listed after indexes finish loading.
  • Agents and Git: Export a thread with tool results, import it in a test environment, and confirm query results and compressed artifacts are present. Exclude a scheduled flow from a managed source and confirm it no longer runs.
  • Configuration: Open Config Profiles on a deployment with a large configuration and confirm the page loads without freezing. Import a connection bundle and confirm secret ownership matches the source environment.
  • Validate after upgrade: Send a test chat message, run a query against each configured connection, and confirm provider status on the overview page for AI providers.
— The Genesis Team