Table of contents
Official Content
  • This documentation is valid for:

Version 2026-01 was officially released on Globant Enerprise AI on March 1, 2026, and includes additional updates delivered in Hot Fix 1 (March 10).

Below are the most important fixes and features introduced in this version and its Hot Fix.

Version 2026-01 Hot Fix 1 (March 10)

  • Security
    • Fixed vulnerabilities in the Teamstudio-Worker container image reported so you reduce the attack surface.
      • Updated base image and dependencies to align with current scanning requirements.
    • Fixed high-severity Java-related vulnerabilities in API, Console, and LLM Server detected in pre-production so you meet security expectations before rollout.
    • Fixed findings for the omni-parser image so you comply with current security scan requirements.
    • Improved container hardening across services so you benefit from scan-based remediation, refreshed base images, and dependency updates that strengthen your security posture.
  • RAG
    • Improved OmniParser document processing reliability so you consistently receive complete section extraction, including content derived from images.
    • Fixed a runtime error in OmniParser when calling /v1/omni-parser/process (“list index out of range”) so you can process affected documents without failures.
  • Flows
    • Fixed authentication for the embedded Flow chatbot so you no longer see unnecessary login prompts or error pages in production.
  • Console
    • Fixed a validation error when assigning permissions to assistants from Roles > Permissions so you can complete permission assignments.
    • Fixed layout and visualization regressions introduced after the 2026-01 release so you can navigate and operate the Console as expected.

Version 2026-01 (March 1)

  • Lab
    • Integrations now appear in the Side Menu, replacing the former Tools option. 
      • Integrations allow you to group Tools so they can share common parameters. For example, multiple Tools in the same group can use the same authentication key. 
      • An automatic migration converted standalone Tools (Tools created without grouping) into Integrations that group one or more Tools. Relevant parameters and credentials were automatically moved to the group level.
      • Integrations support JSON definitions based on OpenAPI and MCP, allowing you to import OpenAPI 3.x, or connect to an MCP Remote Server.
      • Show Tools in Integrations Dashboard, improving The Lab and user experience consistency; fix CSS in Security Scheme, Integration Parameters and Tool Parameters tabs; show built‑in edit constraints; adjust labels/texts and empty states; match The Lab data model to the Integrations API:
      • Compose Tool names with Integrations name; list Integrations with filters; delete Integrations with cascade and soft-delete rules; limit the number of Tools per group; expanded detail level to return Tool and parameter values; publish all draft Tools of Integrations in bulk.
      • Endpoints include full parameter hierarchy and allow inheritance where applicable.
      • For more information about the automatic migration, see Migration from Tools to Integrations.
    • New Credential Manager with list, create, and edit workflows (Name, Integration, Type, Status), showing Authentication Level and Type, and improved error propagation.
      • Credential Manager allows exactly one active credential per integration in the API and The Lab, with clear rules and statuses for activation and deletion.
      • Treat clientId/clientSecret and fromSecret defaults as secrets (masked in The Lab/API), and keep values encrypted at rest.
    • Tools improvements:
      • New SharePoint Tools for reading, listing, and saving files with service account and OAuth support.
        • Read files with dynamic folders and multi-auth.
        • Save files (PDF/Word) to dynamic folders with multi-auth.
        • List folders/files and read document content.
      • New Email Tool capabilities to build dynamic HTML bodies, subjects, recipients, and attach files generated by Agents.
      • New Google Drive Integration action to create PDFs (headings, lists, tables, images).
      • New default API Key fallback for Web Search Tools when no Credential is set (env var CREDENTIAL__APIKEY).
    • New Agent publication checks validate that Integration parameters and credentials are complete before publishing.
    • Agents and Iris improvements:
      • Iris now includes Agent presentation (intro, description, starters, features) and avoids adding process-only tools; running older Agent revisions now works as expected.
    • Files and multimodal playback improvements:
      • Fixed file uploads stuck in antivirus scan and restored audio/video playback in the multimodal player; aligned allowed file categories across views.
    • Fixed multiple Import/Export issues (Agent details mismatch after import, dark banner on light mode, integration import empty response, GetToolPlugin by name/ID).
    • Fixed missing icons and missing public Jira in dashboard.
    • Fixed OAuth-based Tools failing to authenticate and access user documents.
    • Fixed send-email tool PDF generation by selecting a supported pdf-engine.
    • Security sanitized callApi logs to avoid leaking tokens/keys.
  • Station
    • New Favorites model and endpoints across Station and Middleware, allowing you to favorite/unfavorite published Solutions from any project and view them in a dedicated My Favorites section.
      • Favorites list, mark/unmark from cards, pagination, and solution metadata (solutionId, externalVersion/Revision, permissions, subscription).
    • New scoped Station Admin roles and login resolution (Organization Admin and Project Admin) propagated through Station and Middleware.
      • Frontend and backend now resolve admin privileges for the current organization/project context; login/session endpoints include scoped data.
    • Taxonomy administration improvements:
      • Install default taxonomies from editable JSON via Console; migrate all organizations in bulk with safe purge semantics; resolve Translation Agent by Agent Name.
    • Gallery and Workspace UX improvements:
      • Infinite scroll with accurate counters and “You’re all caught up”; preserve filters after login redirect; fix “Explore Solutions” filter behavior; Show Lab Version.Revision in Detail; updated CODA card and logo; reposition Share button; add test-ids; persist selected language across sessions; add 404 page.
    • Improved My Creations to list project Solutions by type, status, and latest Lab vs. Published/Unpublished revisions, with middleware proxy alignment.
    • Fixed light mode visuals and multiple UI regressions (login background height, “Showing results for” hint, proceed/change project flow, caught-up message logic, CODA external image).
    • Deprecated Assistant creation in Console in favor of Agents in the Lab (UI disabled and deprecation banner shown).
  • Console
    • Evolution from Assistants to Agents:  
      • Creation of Chat and API Assistants in the Console has been disabled.
      • Creation through the API remains temporarily available but carries risks because newly created assistants will not be migrated and will be deprecated in the future.
      • Going forward, it is recommended to use Agents.
      • Data Analyst Assistant 2.0 has been discontinued. 
      • For more information, see Evolution from Assistants to Agents.
    • Model Configuration Controls: Organizations can now define which LLMs are enabled, improving governance and cost management.
    • New UI management for organization-approved LLM Providers/Models, including bulk add/remove actions and Assistant editors that filter only approved options.
    • New Usage Dashboard Pivot with Month, Provider, Model, Cost, Requests, Total Tokens and KPIs across project dashboards.
    • Improved file uploads in Console after image hardening and fixed Prompt Files URL formatting (pretty route).
    • User-Id is now stored for Agentic Process executions in The Lab > Jobs. The Owner field in The Lab > Jobs > Subject Details View (General tab) now identifies the user who initiated the execution, improving tracking and auditing.
    • Fixed CSV export security permission issues across multiple grids.
    • Security blocked unsupported HTTP verbs (PUT/DELETE) in Backoffice endpoints.
  • Workspace
    • New feature flag to temporarily disable the new Run flow and re-enable the legacy Run in this release, while keeping navigation in the same tab.
    • Improved deep-linked navigation by restoring all filters and query parameters after login.
    • Improved language persistence so the selected locale survives logout/login.
    • New and improved Pinned/History modules and API responses (pinId, grouping, extended Station fields).
  • API
    • New Integrations API to register, retrieve, update, and delete Integrations, including security scheme management and import/export operations.
    • New Organization scope for Public Providers so only models explicitly allowed to your organization appear across APIs (LLM API, Chat API).
    • New Organization API endpoints to bulk Delete model overrides by provider or reset all overrides for an organization.
    • New Access Control API endpoint to remove a user from a Project.
    • New Terms & Conditions APIs to check and register acceptance per organization/installation.
    • New File API support for uploads coming from Chat (up to 1 GB, with type/size validation).
    • Improved error transparency for LiteLLM errors and authorization failures.
      • Return descriptive error bodies for Chat/LLM calls instead of “LiteLLM General Error” when safe to display.
      • Return meaningful 401/403 JSON errors on GET /accessControl/organization/plugin-runtime-policies.
    • Improved stability by separating HTTP timeouts (connect/socket/pool) to avoid hanging connections.
    • Improved Usage Limits API and role checks to avoid 404 “Project not found” for Provisioning_Services role.
    • Fixed Chat API to support POST only (harden HTTP methods).
    • New MCP runtime to execute Tools over remote MCP with user-based authentication.
  • Flows
    • Flows Conversation History will be deprecated and replaced by a new section on the platform to display records.
    • New OAuth login support for GEAI-managed channels (Web Chat), including token validation and backend permission checks; the Chat API now forwards the GAM token to Flows via x-saia-access-token.
    • Improved Spring Boot and Keycloak dependencies to current supported versions for security hardening and compatibility.
      • Spring Boot upgraded to 3.x.
      • Keycloak SDK upgraded to 25.0.5.
    • Improved database portability by migrating Flows storage from MongoDB to CosmosDB for managed-service readiness.
    • Improved Content Security Policy configuration using an environment variable for custom deployments.
    • Fixed Slack interactive button events not handled when translating messages to Slack.
    • Fixed partial responses not shown in Web Chat after a second message.
    • Security enforced OAuth flow guardrails and error handling for unsupported channels.
  • RAG
    • New LLM ingestion strategy for omni‑parser to process entire pages with the configured LLM when hi_res OCR yields misreads.
    • Improved multimodal embeddings by honoring chat.ingestion.multimodalBatchSize for providers that require single‑item batching.
    • Fixed PPTX/XLSX ingestion rejections from Console uploads.
    • Security removed sensitive payload elements from omni‑parser/RAG logs and addressed Trivy findings.
  • LLMs
    • New Anthropic Claude 4.6 Opus and Sonnet across providers. Incorporation of Claude Opus 4.6 and Claude Sonnet 4.6, Anthropic’s flagship models optimized for advanced coding and agentic, multi-step workflows with tool use. Available now in production via Anthropic, and coming soon via AWS Bedrock and Google Cloud Vertex AI:
      • anthropic/claude-opus-4-6 — Tool use, streaming, multimodal.
      • vertex_ai/claude-opus-4-6; awsbedrock/us.anthropic.claude-opus-4-6-v1 (+fallback to Vertex AI).
      • anthropic/claude-sonnet-4-6 and cross‑provider fallbacks to Vertex AI.
    • New Google Gemini models. Integration of Gemini 3.1 Pro Preview, Google's latest reasoning model featuring a 2x+ improvement in reasoning performance over Gemini 3 Pro, enhanced token efficiency, and optimized agentic multi-step workflows with precise tool use. Available in preview via Google Cloud Vertex AI:
      • vertex_ai/gemini-3.1-pro-preview (multimodal, streaming, tool use).
      • gemini-3-flash-preview (Vertex AI + Gemini API).
      • vertex_ai/gemini-3-pro-preview.
    • Integration of GPT-5.2-Codex, an upgraded GPT-5.2 variant optimized for agentic coding tasks, available via the Chat API or the Responses API:
      • openai/gpt-5.3-codex.
      • openai/gpt-5.1-codex, gpt-5.1-codex-mini, and gpt-5.1-codex-max.
      • Deprecated codex-mini-latest migrated to openai/gpt-5.1-codex-mini. For more details on the migration, please refer to Deprecated Models.
    • New Image generation models:
      • openai/gpt-image-1.5 and gpt-image-1-mini.
    • New providers/models:
      • vertex_ai/zai-org-glm-5-maas; OpenRouter GLM‑4.7; Nvidia DGX Nemotron‑49B‑v1_5; OpenRouter MiniMax M2.1; OpenRouter DeepSeek V3.2 family; Globant DGX GLM‑4.6.
    • New Azure AI Foundry additions including gpt‑5.x family and Model Router support for dynamic model selection.
    • Improved Bedrock - Vertex fallbacks for Claude models to mitigate rate limits and account restrictions.
    • Fixed LiteLLM container image issues (remove test keys, address missing vertexai import, reduce CVEs) and restored OpenAI gpt‑5 in remote module for new installs.
    • New model optimized for coding and agent-based workflows:
      • openrouter/minimax/minimax-m2.1
    • New Qwen3-VL Models added:
      • openrouter/qwen3-vl-235b-a22b-instruct
      • openrouter/qwen3-vl-235b-a22b-thinking
      • openrouter/qwen3-vl-30b-a3b-instruct
      • openrouter/qwen3-vl-30b-a3b-thinking
      • openrouter/qwen3-vl-8b-instruct
    • Fixed LiteLLM General Error issues in Assistants/Chat with clearer messages and recommended configuration (e.g., token limits, alternative models).
  • Security
    • Security hardened container images across API, Console, Lab, RAG/omni‑parser, and LiteLLM (non‑root execution, removed shells/tools, labeled builds, addressed CVEs).
    • Security improved Chat/LLM error redaction while preserving actionable messages, removed sensitive data from logs, and masked fromSecret defaults in APIs/UI.
    • Security restricted Chat API to POST, tightened method handling in Console, and fixed multiple OWASP/SBOM findings.
    • Security enforced organization-level LLM enablement at runtime (direct calls, Agents, Assistants).
  • Known Issues
    • Some third‑party models may present provider‑side deprecations or quota constraints. Fallback models are configured where possible. If a model fails due to provider limitations, select an approved alternative from your organization’s allowed list.

Required manual steps

Last update: December 2025 | © GeneXus. All rights reserved. GeneXus Powered by Globant