Skip to main content

​
v1.10.4 — March 6, 2026

​
On-demand tools and context management, 3-tier model selection, new connectors, RBAC enforcement, concurrency scaling, and broad reliability hardening

​
Context-Aware Tools and Skills

Tools, skills, and context are now managed in a new, more scalable and token-economical way. Agents search for and load capabilities on demand rather than carrying every tool in every request.
  • Agents discover and activate tools as needed instead of loading all tools upfront, significantly reducing token usage per request
  • Default tools and skills can be pinned as “always on” per agent for capabilities needed on nearly every turn
  • If an agent doesn’t think it can do something it should, nudge it to “search your tools context” — it will find and load the right tool
  • Context-control service manages tool/skill budgets, thread guardrails, and runtime introspection
  • Context-tool call activity is now visible in the chat UI by default for transparency

​
3-Tier Model Selection

A new 3-tier model structure gives users and admins fine-grained control over which model handles each interaction.
  • Three model tiers — Premiere (frontier), Normal (main), and Fast — available across all supported LLM providers
  • Switch tiers on the fly from the wrench icon in the upper right of the chat UI
  • Set different default tiers per agent and per blueprint phase for cost/quality trade-offs
  • Model selections have been reset to Premiere: Opus 4.6, Normal: Sonnet 4.6, Fast: Haiku 4.5
  • Per-thread and per-message tier overrides for targeted quality control

​
New Connectors and Integrations

New enterprise data source connectors were added, expanding Genesis’s reach into collaboration and metadata platforms.
  • Add Confluence connector with cloud and data center support, including spaces, pages, and attachment operations
  • Add SharePoint connector with list and download via delegated user OAuth
  • Add SharePoint client-credentials auth for tenant-wide app-only access
  • Add OpenMetadata connector tool with full CRUD, search, and egress-aware endpoint handling
  • Add custom Bedrock endpoint support with configurable auth header for enterprise API gateways

​
RBAC and Access Governance

Access control moved from advisory checks to enforced dual-key verification across all agent tools and connection paths.
  • Enforce dual-key RBAC across all agent tools — both user and bot must have access
  • RBAC single-key default with opt-in double-key per connection
  • Remove dynamic owner roles; resolve ownership from metadata instead
  • Filter per-object access grants from user permissions response
  • Remove budget fallback polling and speed up RBAC user lookups on Snowflake
  • Agent-to-app RBAC access inheritance — apps inherit access from creating agent

​
Concurrency and Performance

Targeted fixes eliminated bottlenecks that previously limited the platform to ~10 concurrent users.
  • Concurrency scaling fixes — 0 errors at 100 concurrent users
  • Fix SQLite write contention in cost/usage metrics service with dedicated writer thread
  • Add GIL-yield points in schema explorer to reduce event-loop lag
  • Offload plist-backed WebSocket history reads to background threads
  • Cache blueprint registry to eliminate redundant YAML reads
  • Express-lane tool-transition optimizations and async title generation
  • Enhanced tool chain timing with in-flight tracking and model-gap breakdown

​
Genesis Scanner Pipeline (Digital Twin)

A full Genesis object scanner was built from scratch, enabling the Digital Twin to index all internal Genesis objects and their relationships.
  • Add Genesis scanner core module with SQLite database scanning
  • Add dashboard and dataset metadata nodes with data flow edges and SQL lineage
  • Add schedule nodes and object relationship edges to the scanner pipeline
  • Add message-log mining for semantic enrichment and provenance edges
  • Add microapp scanning with LLM-based Python code analysis
  • Add internal genesis-scanner HTTP endpoints for server-side computation
  • Cleanup dead message_log_miner code and fix microapp node type mapping

​
Snowflake and Deployment Operations

Release and deployment workflows gained safer controls, improved environment handling, and operator visibility.
  • Automate changelog generation on release branch creation
  • Split release flow into create-release and create-rc orchestration
  • Add check-engine indicator lights to release dashboard
  • Make release-dashboard check indicators data-driven and add UAT pill
  • Add per-environment UPDATE deploy controls to release dashboard
  • Dispatch Snowflake GA deploy alongside Railway on promote-to-ga
  • Unify platform deployment config for multi-platform support
  • Cleanup: remove obsolete Snowflake workflows, sync oauth-proxy config
  • Add Slack notifications to Snowflake deploy workflow

​
Databricks and EKS

Enterprise deployment paths were expanded for Databricks persistence and AWS EKS hostname management.
  • Add Databricks persistent storage with snapshot/restore across deploy cycles
  • Support permanent EKS hostnames via configurable host variables
  • Add update-eks-image script to upgrade all three EKS deployments
  • Automate customer onboarding — ECR registry and license key generation

​
Chat, Handoff, and UI Reliability

Thread continuity and user experience were hardened across chat, notifications, theme, and panel behavior.
  • Harden chat, budgets, and microapp runtime recovery
  • Preserve original URL after Auth0/OIDC login redirect
  • Skip thread ownership check for synthetic “adhoc” thread ID
  • Avoid redundant thread title lookups in thread list
  • Restore one-on-one title updates and subscribe bot_id
  • Close right-side panels on thread switch to prevent cross-thread bleed
  • Fix sandbox following user to unrelated chat threads
  • Harden system theme synchronization for automatic dark/light mode
  • Improve dashboard/app organization with tag-based filtering in the Apps view
  • Fix AG-Grid Databricks row grouping showing field names instead of values
  • Handle HTTP 202 in SQL API async query submit and status check
  • Harden request throttling and thread history handling for multi-tab traffic
  • Respect backend rate-limit windows in WebSocket reconnect flows

​
Git Browser and Blueprint Workflows

Git and blueprint workflows gained safer defaults and more resilient execution paths.
  • Harden git tooling and blueprint managed-repo workflows
  • Allow mission start when blueprint sync fails — show warning instead of blocking
  • Improve blueprint creation prompts for stronger phase exit criteria
  • Relax git policy for automated branch work

​
Tools, Skills, and Platform Extensibility

Agent tooling and operational capabilities were expanded for runtime administration and observability.
  • Strengthen sandbox build/start reliability and live work log refresh
  • Expand support docs indexing and startup diagnostics
  • Suspend metrics bridge uploads after consecutive DNS failures
  • Harden scheduler SQLite recovery and blueprint index cache invalidation
  • Pin APScheduler before v4 and guard next_run_time access
  • Mission initiator targeting for correct user attribution
  • Replace oauth-proxy deployments symlink with real file
  • Show context-tool calls by default and add execution lane to LLM logs
  • Harden tool_call and Snowpark runtime handling

​
Internationalization

  • Add Spanish i18n foundation with translations across all major dashboard surfaces

​
Upgrade Notes

Upgrade from v1.9.2 to v1.10.4 includes enforced dual-key RBAC, new connector integrations, and significant concurrency/performance changes. Validate role mappings, connection access policies, deployment defaults, and environment configuration in lower environments before production rollout.