Genesis v1.19.49 — May 22, 2026
Repo-synced Apps, Skills, and Flows; Microsoft Teams Bot Framework; admin audit logging; chat reliability; and broad platform improvements across connections, blueprints, and agent tooling
Genesis Apps, Skills, and Flows: Repository Sync and Managed Artifacts
Apps, Skills, and Flows can now be managed as first-class repository-synced artifacts with a uniform dependency model across all three surfaces.- Skills, Flows, and Genesis Apps share a consistent managed-source workflow: operators can bind artifacts to Git repositories, track dependencies, and sync changes through the same multi-instance sync machinery used for blueprints.
- Managed-source configuration on Apps, Skills, and Flows is now tucked behind a modal, keeping day-to-day editing screens uncluttered while still exposing full sync controls when needed.
- Export-to-commit and sync-from-repository operations are unblocked for app management workflows that previously stalled on repository state checks.
- This means you can keep your Skills, Flows and Apps artifacts in your own code repository
- Genesis Apps that run for extended sessions no longer lose authentication after six hours
- Agents working on versioned Apps are no longer nudged to auto-promote production copies; promotion remains an explicit operator action.
Microsoft Teams: Bot Framework and File Attachments
Microsoft Teams integration expanded beyond webhook notifications to full conversational Bot Framework support.- Genesis agents can now participate in Teams conversations through the Microsoft Bot Framework, enabling threaded back-and-forth directly in Teams channels and chats — not just outbound alerts.
- Agents can send file attachments in Teams outbound messages, so reports, exports, and generated documents arrive as downloadable files rather than inline text.
RBAC, Audit Logging, and Access Governance
Administrators gained stronger visibility and finer-grained controls over who can do what.- An admin audit log now records user, role, policy, and configuration changes, giving compliance teams a traceable history of governance actions.
- The role editor includes a live access preview: operators can see exactly which connections, Apps, and resources a role grants before saving.
- App-tag bulk access assignment UX was improved so granting a permission level to all Apps sharing a tag is faster and less error-prone.
Connections and AI Providers
The Connections experience was streamlined and several new provider types and reliability fixes landed.- The Connections page now uses a single header search bar instead of separate per-tab filters, making it faster to find a connection across all types.
- A new Astronomer connection type for Airflow and Astronomer-hosted orchestration workflows.
- LiteLLM Proxy connections can be configured through Helm with OAuth credentials, and proxy header reconfiguration and provider card state are now reliable across restarts.
- PrivateLink-aware URL handling prevents broken OAuth and API calls when Genesis runs behind a Snowflake PrivateLink endpoint.
- The connections tool now exposes which roles have access to each connection, helping agents and operators understand permission boundaries.
- OpenAI model defaults were updated to GPT-5.5 for new deployments.
- Anthropic prompt caching defaults were repaired so cache-eligible calls benefit from reduced cost and latency as intended.
Chat, Threads, and the Sidebar
Several fixes eliminate visible glitches and add useful organization options in the chat experience.- Switching between threads no longer leaves duplicate streaming segments, corrupted sender labels, or stale tool-call rows in the transcript.
- Tool-call progress narrative rows are restored during streaming, so operators can follow multi-step agent work in real time without gaps.
- A “Group by people” toggle in the chat sidebar lets users organize threads by participant rather than only by recency.
- Thread visibility in the sidebar and the long-running completion overlay were hardened so threads do not disappear prematurely and completion timers reflect actual elapsed time.
- Interactive credential cards and secret prompts are dismissed automatically when the user sends a new message, keeping conversations uncluttered.
Context Graph and Agent Knowledge
The unified wiki and Context Graph viewer received reliability and recall improvements.- The LLM wiki and Context Graph wiki layers are now merged into a single user-facing wiki, eliminating duplicate surfaces and inconsistent retrieval paths.
- Agents can recall past SQL queries by the tables and views they touched, with improved anchoring to validated query history.
Blueprints and Missions
Blueprint and mission workflows gained API support, safer exit checks, and operator-quality-of-life improvements.- A BlueprintRun API enables programmatic blueprint execution and external demo orchestration, with hardened idle-nudge behavior when a blueprint waits for operator input.
- Missions can now be renamed from the dashboard without recreating them.
- Mission list filters remain sticky across navigation, so operators do not lose their filter context when switching pages.
- Long blueprint descriptions in the Start Mission modal are clamped so the dialog remains usable.
Git Browser and Repository Export
- Dotfiles (such as
.gitignoreand.env.example) are now visible in the Git browser, so agents and operators can inspect and edit configuration files that were previously hidden.
Agent Tools: Excel, Python, and State Verification
Agent-executable tools received capability and safety upgrades.- A new Excel tool group supports structured read and write operations on spreadsheet files, enabling agents to produce and consume Excel deliverables without manual file handling.
- Python execution reliability was improved with sandbox guards, judge retry logic, and correct genesis SDK resolution inside Docker — agents are no longer misdirected to incorrect package sources when the SDK import fails.
- Agents now verify external system state before mutating shared resources (databases, repos, ticketing systems), reducing accidental overwrites when concurrent changes occur.
SCIM Provisioning
- SCIM provisioning now ingests Okta enterprise user attributes, keeping Genesis user profiles aligned with directory fields such as department and cost center.
Word Export
- Mermaid diagrams and Plotly charts embedded in agent responses are now rendered as images in Word document exports, so exported reports display visuals correctly rather than raw markup.
Support and Observability
- Thread exports now include runtime build metadata, giving support teams version and configuration context alongside conversation content.
- LLM call logs include finish reason and configured output token cap, making it easier to distinguish natural completions from token-limit truncations during triage.