Skip to main content

​
Genesis v1.19.84 Release Notes

July 23, 2026 Launchpad onboarding and estate measurement for migration validation, stronger agent tooling and handoff continuity, and reliability improvements across demos, AI providers, and the dashboard.

​
Launchpad and Onboarding

New guided onboarding helps every user get productive faster, with clearer migration planning and repeatable sales-demo journeys.
  • Launchpad onboarding adds a first-login tour and a state-derived setup checklist so new users know exactly what to configure next.
  • Each user can choose their own default landing page instead of always opening Launchpad.
  • The migration wizard now shows source and target connections side by side for easier comparison before starting a migration.
  • Sales Engineering demos use standard product journeys instead of private demo cards, with repeatable client-file onboarding scenarios.
  • The onboarding checklist can advance memory-related setup steps without getting stuck on missing tools.

​
Estate Measurement and Migration Validation

Teams can characterize a database estate, freeze a point-in-time snapshot, and compare estates to validate migrations with evidence.
  • Agents can measure a database estate, freeze it, and compare two frozen estates directly from chat.
  • Source-anchored measurement characterizes an estate, self-verifies the result, and produces signed frozen contracts for diffing.
  • An in-app approval and verification UI for transition contracts lets reviewers sign off on migration readiness without leaving Genesis.
  • Composite, text, and human-designated row-identity keys support migration verification when simple column keys are not enough.
  • Measurement now covers timestamp, decimal, and boolean columns instead of reporting them as unmeasurable.
  • Databricks integer and boolean columns are compared by type instead of as text, reducing false positives.
  • Databricks NULL timestamps no longer appear as differences when the underlying data matches.
  • A frozen contract’s content hash is preserved on load instead of being silently rewritten.
  • Measurement passes survive a transient statement failure instead of aborting the entire estate scan.
  • A schema-prefix migration no longer returns a misleading verdict when it compared zero objects.
  • An estate dashboard provides a visual front end for estate characterization and comparison results.
  • Opt-in retention and compaction for the System Diff reconciliation store helps long-running validation programs stay manageable.

​
AI Providers and LLM

OpenAI routing, provider configuration, and blueprint reliability are improved for more predictable agent behavior.
  • OpenAI now defaults to the LiteLLM engine, aligning OpenAI with the same routing stack used for other providers.
  • One-shot completions stay on the proxy protocol so structured output (response_format) is enforced consistently.
  • Rate-limited blueprint missions stop immediately instead of waiting for a timeout.
  • Provider configuration resolves once per message for faster and safer concurrent dispatch.
  • Operators can set or clear the OpenAI base URL without activation silently baking the environment endpoint into stored config.
  • Unroutable model strings are rejected at the assignments boundary instead of failing at call time.
  • Embedding credentials resolve from App Config instead of a stale environment alias.
  • Blueprint thread creation with Claude on Cortex and LiteLLM no longer fails on thread-based creation paths.
  • Chat completion errors return typed failures instead of opaque error strings buried in response text.

​
Demo and Sales Engineering

Shared demo environments are safer, and presenters get tools to reset scenarios and run polished demonstrations.
  • A presenter panel prepares, resets, and cleans up demo scenarios so every presentation starts from a known state.
  • Presenter controls work for staff behind an OAuth proxy by trusting proxy-forwarded email headers.
  • Shared demo environments fail closed and isolate concurrent Sales Engineer runs so one session cannot corrupt another.
  • Reusable blueprints stay out of disposable demo scope so shared assets survive cleanup.
  • Lab redeploys no longer break secret resolution, Databricks checks, or Slack setup.
  • Sales Engineers can run deterministic demos for an AUM app, cost optimization with approval gates, and dashboard correction with PR review from natural-language prompts.
  • Verified Flow RCA completion is preserved when follow-up model responses fail, so investigation results are not lost mid-demo.
  • Presenters can simulate a pipeline failure and watch Eve investigate it automatically in Slack.

​
Chat, Handoffs, and Threads

Handoff workflows, compression, and thread navigation are more visible and reliable.
  • The chain strip shows a live running indicator, and a follow-the-handoff toast appears when work moves to a child thread.
  • Handoff toasts stay visible until acted on and dismiss on click or navigation.
  • Handoff completion notifications no longer fire prematurely for blueprint-chain children or nudge notify-mode children to chain after storing outputs.
  • Completion notify resolves by child thread ID when the handoff ID property is clobbered.
  • Background summary cache warming pre-compresses thread summaries during idle time so long threads open faster.
  • Compression history shows plain-language summary-warming rows so users understand what was pre-computed.
  • The new-chat control stays available when threads are grouped by user.
  • The chat token-usage summary adds a Total tokens line and reconciles with the pricing breakdown; cached tokens are no longer double-counted in the cost panel.

​
Agent Tools and Connections

Agents gain broader data access, safer multi-tenant isolation, and more capable file and Git workflows.
  • execute_python_code scripts can write files into the connected Git repository.
  • connection_api_action can reference on-disk files, and the file tool uses 1-based line numbers for clearer reads.
  • Large truncated API reads recover via an on-disk file instead of a fragile base64 decode path.
  • max_rows_override expands the LLM-visible query preview beyond the default 20-row cap (up to 500 rows).
  • A generalized SQL validator works across dialects, replacing the Databricks-only alias.
  • The wait tool is available in the always-on context tool group so every agent can pause when needed.
  • find_past_queries no longer fires on non-SQL artifact requests.
  • Agents can copy and move Databricks workspace notebooks or files across connections without manual downloads.
  • git_action push and pull no longer fail with “no upstream branch” on newly created branches.
  • Concurrent Snowflake agents can no longer query another agent’s database or schema.
  • Agents can choose public or internal Jira Service Management comments without changing existing behavior.

​
Missions and Dashboard

Mission workflows and the dashboard UI are more stable and easier to navigate.
  • The mission overview no longer crashes with an infinite render loop (black screen).
  • Mission overview dividers support hover and corner resize; re-clicking a sidebar section returns to that section’s root.
  • Returning to a background browser tab reconnects WebSockets and rechecks server health.
  • Large document index listings are paginated so agents and the UI can reach every file.
  • External-access approval no longer shifts or clips the Genesis workspace layout.

​
Slack

  • Switching the LLM provider no longer kills the shared Slack worker that blocks all Slack sends.

​
Scheduling

  • Scheduled jobs support a relative one-shot trigger (run_in_seconds / run_in_minutes) for near-term automation without crafting absolute timestamps.

​
Security

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

​
Deployment Artifacts

Use these exact references when deploying this release:
  • Genesis release: v1.19.84 (promoted from v1.19.84-rc.11)
  • Docker Hub image: genesiscomputingai/genesis:v1.19.84
  • Docker Hub digest: genesiscomputingai/genesis@sha256:e4fefb0cf7e5617aa670be31af49000a659c3108aa3133c825530d6d794d2d96
  • Helm chart: oci://registry-1.docker.io/genesiscomputingai/genesis
  • Helm chart version: 0.3.63 (appVersion: v1.19.84)

​
Upgrade Notes

  • Launchpad onboarding: Sign in as a new user and confirm the first-login tour and setup checklist appear. Change your default landing page in profile settings and confirm the next login opens the chosen page.
  • Estate measurement: From chat, measure a database estate, freeze it, and compare two frozen estates. Confirm timestamp, decimal, and boolean columns appear in results. Open the estate dashboard and verify characterization and comparison views load.
  • Migration validation: Run a transition contract through the in-app approval UI and confirm reviewers can sign off. Test a schema-prefix migration and verify the object count in the verdict matches expectations.
  • AI Providers: Confirm OpenAI chat and structured-output calls succeed through the LiteLLM engine. Set and clear the OpenAI base URL and verify it persists across re-activation. Trigger a rate-limited blueprint and confirm it stops promptly.
  • Demos: Open the presenter panel, reset a demo scenario, and confirm a fresh state. Run two concurrent Sales Engineer sessions and verify they do not interfere.
  • Chat and handoffs: Start a blueprint chain handoff and confirm the chain strip and follow-the-handoff toast appear. Open a long thread and check compression history for summary-warming rows.
  • Agent tools: Run a Python script that writes to the Git repo and confirm the file appears. Read a large query result with max_rows_override and confirm more than 20 rows return. Paginate through a large document index listing.
  • Validate after upgrade: Send a test chat message, run a short estate measurement, and confirm AI provider status on the providers overview page.
— The Genesis Team