Glob.AI OS Version 2026-03 was officially released on May 6, 2026 and includes additional updates delivered in Hot Fix 1 (May 14), Hot Fix 2 (May 15), Hot Fix 3 (May 27), Hot Fix 4 (May 27), Hot Fix 5 (May 28) and Hot Fix 6 (May 29).
No work items released in this Hot Fix.
No work items released in this Hot Fix.
- Agentic Hub -Run
- Fixed intermittent missing attachments in Workspace conversations, so you see uploaded files render consistently across new and existing threads.
- Caching for plugin metadata now keys by pluginId, pluginVersion, and pluginRevision to prevent cross-version reuse.
- Behavior validated for conversation lists, recent conversations, and conversation details.
- Agentic Hub
- Improved the Solution Detail page layout to align with the latest design system and improve content organization.
- Separated the header section from the description and content areas.
- Added a dedicated header block for avatar, name, role, rating, and primary actions.
- Added a dedicated content block for description, documentation, and additional metadata.
- Added a separate reviews section to improve readability and visual hierarchy.
- Maintained responsive behavior for smaller screen sizes.
- Fixed missing file upload in OS Run, so published solutions honor the Agent’s “allow uploads” configuration and you can provide files when running a solution.
- Fixed shared conversation links returning 404, so you can open public share URLs and see the correct page.
- Run
- Fixed a regression that prevented the chat interface from rendering when you started a new conversation, so you see the chat immediately without refreshing or switching threads.
- Fixed Drag & Drop overlay interruptions over the input box, so you can drag files anywhere in the chat container and still trigger uploads (including when dropping on the input).
- Improved the conversation banner for outdated or unpublished solutions, so you get accurate messaging when a newer version replaced the solution and the conversation is read-only for reference.
- Fixed shared conversations so anonymous viewers can see images in the chat history, enabling complete visualization when you share a conversation.
- API
- Improved Conversation APIs so you can determine file upload availability without extra calls. Conversations now include plugin upload capability fields when you list or create them.
- Added fields in responses: pluginUploadEnabled, pluginUploadMaxFileCount, pluginUploadMaxFileSize, pluginSupportsRawHtml, pluginSupportedFileExtensions.
- Covered endpoints: GET /v1/conversations, GET /v1/conversations/recent, GET /v1/conversations/open (and any list endpoint).
- When pluginUploadEnabled = true, you can render upload controls immediately and enforce the provided limits.
- Fixed POST /images/edits returning empty URLs, so you can retrieve generated images reliably.
- azure_ai_foundry/FLUX.2-pro now returns a valid, accessible URL.
- File retrieval works via File API: GET /v1/files/{fileId} and GET /v1/files/{fileId}/content.
- Images render correctly in Workspace and can be served to end users from Flows.
- Fixed a failure when you called the Chat API with guardrails enabled that returned 500 “Unexpected end of JSON input” instead of a valid response.
- The issue was reproducible with llm-output-guardrail and affected QA and Beta environments.
- Response assembly and JSON serialization were hardened so the API always returns a well‑formed JSON object.
- Tools/Agents
- Fixed image generation with Grok 2 so you can view results in chat. Generated images are now downloaded, stored in the configured storage, and returned with a renderable reference for the Chat UI.
- RAG
- Improved Pinecone capacity and scalability by migrating the production indices to the serverless “sprod” index. You benefit from elastic capacity and consumption-based billing with no fixed POD limits.
- Your existing RAG Assistants were repointed to the new index without changes to prompts or metadata.
- Legacy index are being retired after stability monitoring to ensure continuity.
- Agentic Hub
- Improved solution discovery reliability. You no longer experience timeouts in Explore when loading the Solutions list because popularity/trending scores are now computed asynchronously once per day and persisted for reuse.
- Additional technical logs help you trace async job start/end, persistence success, and fallback usage.
- Intelligence Hub
- New auto-injected prompt caching checkpoints. You benefit from automatic cache-control injection at the system message level across Anthropic, Vertex AI (Claude and Gemini), Amazon Bedrock, OpenRouter, and Azure AI Foundry (Claude), reducing latency and cost on cache hits with no code changes.
- Centralized configuration via cache_control_injection_points.
- Consolidated Vertex AI Gemini preview entries under a single vertex_ai/gemini preview pattern.
- Improved pricing accuracy with cost schema v2. You get correct billing for cache reads/writes and modality-specific inputs across providers.
- Adds cache read/write pricing to Anthropic, AWS Bedrock (Anthropic), Google Vertex AI (Claude; Gemini read-only via Cache API), Azure AI Foundry (Claude; OpenAI parity with v2), Azure OpenAI (cache read only), and xAI.
- Adds audio_tokens pricing for eligible Gemini models in Vertex AI. Applies provider-specific token thresholds (for example, 200K or 272K tokens) where required.
- Fixed OpenAI cache-hit overcharging in model pricing. You are now billed correctly when OpenAI returns cached prompt tokens because active Chat models were migrated to the version 2 schema and equipped with cached_tokens rates.
- Most OpenAI Chat models moved from v1 to v2 and now include prompt_tokens_details.cached_tokens pricing; tiered thresholds are handled where applicable.
- Excludes models without cached pricing and soon-to-be-deprecated previews as documented in the issue.
- Security updates for container dependencies.
- Improved cost handling based on tokens consumed and cached tokens. You get accurate, condition-based pricing that differentiates regular prompt tokens, cached tokens, audio tokens (where supported), and completions across key model families.
- Anthropic (Claude 4.x variants), Google Vertex AI (Claude via Vertex; Gemini 2.5/3.x), and AWS Bedrock (Anthropic) include cache-aware and tiered input pricing where applicable.
- OpenAI GPT‑5.x models include tiered pricing and cached_tokens handling according to the model’s context thresholds.
- Review provider documentation for region surcharges, thresholds, and cache read/write rules.
- RAG
- Improved scalability and cost efficiency for production vector search. You now use a serverless Pinecone index (sprod) instead of the POD-based prod index, removing capacity limits and adopting pay-per-use pricing.
- Existing assistants were pointed to sprod, and the legacy index is scheduled for decommissioning after verification. No downtime or action is required on your side.
- Console
- Fixed cost reporting for OpenAI and Anthropic (effective from April 1, 2026). You must execute a one-time post-deploy correction to ensure dashboards and reports reflect accurate model costs.
- Agentic Hub
- When you create a new Organization in Console, a Default Project is automatically created and pre-populated with a set of ready-to-use Agents. These Agents are also published to the Solutions so that all users in your Organization can discover and run them immediately. The following Agents are included by default:
- Deep analysis: Generates comprehensive, evidence-based insights from multi-source exploration and synthesis.
- Data Collector: Collects and refines information from the web and scientific sources (ArXiv, PubMed, etc.) for in-depth research.
- Web Search: Retrieves concise summaries and reliable insights from real-time web searches.
- Web page builder: Designs and generates static, informative web pages using HTML, CSS, and JavaScript.
- Social media content creator: Produces engaging, platform-tailored social media posts.
- Legal contract analyzer: Analyzes and summarizes contracts to identify key clauses, dates, risks, and obligations.
- Marketing manager: Provides strategic marketing insights, creative ideas, and brand-consistent content for campaigns.
- Product description generator: Generates SEO-optimized product descriptions to improve visibility and ranking.
- Chat with Popular LLMs: Handles general-purpose tasks such as Q&A, recommendations, and insights. Includes:
- Chat with GPT-5.5
- Chat with Gemini 3.1-pro
- Chat with Claude Sonnet 4.6
- Chat with Grok 4
- Fixed JSON-based Integration import failures after file upload. You can now import valid Integration definitions from JSON files without errors.
- Resolved a 404 error during post-upload processing so the Integration is created successfully.
- Console
- You can now create Organizations in Console or API with the Default Project's base Agents fully saved and automatically published to Solutions.
- Evolution to Glob.AI OS
- As of version 2026-03, the platform is officially named Glob.AI OS.
- This is not just a rename, but a transition to a unified platform designed to support you in the full lifecycle of enterprise AI adoption.
- The goal is to simplify how AI Models are configured, enterprise data is connected, and Agentic Solutions are orchestrated within a single, cohesive environment.
- Glob.AI OS enhances the way in which you can interact with AI solutions across the whole AI-adoption process by clearly indicating where to perform tasks and distinguishing users' responsibilities.
- This evolution represents a shift toward an integrated, hub-based architecture, where the entire platform will be accessed through a single URL, and you’ll see the hubs and options available to you based on your permissions.
- Hubs description
- Each hub has a specific purpose:
- Agentic Hub: Design and orchestrate Agents and workflows.
- Corporate Hub: Integrate enterprise systems, data, and knowledge.
- Intelligence Hub: Configure AI providers, models, and governance.
- The available hubs and options are displayed based on your permissions, allowing you to select the ones to work with.
- What should you know if you were using versions prior to 2026-03?
- As of version 2026-03:
- The platform supports social authentication, enabling you to log in through Google and Microsoft accounts.
- Accessing the Lab and/or Station URLs automatically redirects you to Glob.AI OS within the Agentic Hub.
- No changes were introduced to the Workspace, Console, and/or API URLs in version 2026-03.
- When working within Glob.AI OS.
- You will have access to the Console from the Agentic Hub under the Manage menu (and from the Console, you can access Glob.AI OS).
- If you are an Admin, from the Agentic Hub, under the Manage menu, you can select Solutions, and an admin panel will open to approve, reject, and manage changes to existing solutions.
- Refer to the following table to locate, in Glob.AI OS (version 2026-03), the features formerly available in Globant Enterprise AI (version 2026-02 or earlier).
| Versions prior to 2026-03 |
As of version 2026-03 |
| Agents, Agentic Processes and Flows creation and orchestration were performed in The Lab. |
These agentic solutions are now centralized in the Agentic Hub (Build option in the side menu). |
| Integrations were configured in The Lab. |
Integrations are now available not only in the Corporate Hub, but also in the Agentic Hub (Build option in the side menu). |
| The published AI Solutions were available in The Station. |
They are now available in the Agentic Hub (Discover option in the side menu). |
| Configuration, observability options, and other global settings were managed in the Console. |
The Console is available in the Agentic Hub (Manage option in the side menu). |
| Workspace provided an end users' interface to interact with Assistants and Agents. |
Workspace is considered legacy and will be deprecated soon. It will remain available for a limited time to run existing Assistants defined in your Projects. During this period, it will remain accessible at the same URL.
In Glob.AI OS, when working in the Agentic Hub, you can select Run from the side menu to open a new integrated interface that lets you run the AI Solutions defined there. |
| RAG Assistants were created in the Console and run through the Workspace. This functionality will remain available for a limited time. |
You can find RAG Assistants previously created in the Console in the Agentic Hub (My Creations option in the side menu, Build section).
Additionally, to use a RAG Assistant in the Agentic Hub, you can add it as a Tool so that it is visible and available to an Agent. |
| Globant Enterprise AI provided various APIs that allowed integration and modification of the platform's metadata. |
API endpoints remain available in the same base URLs. |
- Agentic Hub
- My Collections added. It allows users to organize and access their favorite AI solutions in personalized collections, making it easier to group, revisit, and manage solutions based on their needs or workflows.
- Filter added in the Gallery to “Order By” top rated, recently added and popular (most executed ones).
- Explore added. Discover new AI solutions more easily with improved recommendations, enhanced navigation, and a more consistent browsing experience across Explore.
- My Space added. My Space now offers a more personalized experience with quicker access to your favorite solutions, collections, and recent activity.
- Drag & Drop file upload enabled in the Run section
- Improved gallery search and filtering. As a result, you use a left-aligned layout consistently across Discover, My Collections, Manage, and Admin. Explore remains unchanged.
- Improved disabled Run and empty-state guidance by replacing “Lab” references with “Build.” As a result, you receive correct navigation and wording.
- Fixed a persistent success or failure toast that stayed on screen while navigating away.
- Fixed a 404 page that remained visible after being redirected to a valid route when authenticated.
- Improved role-based visibility so you only see solutions allowed by custom roles in Discover, Organization lists, My Favorites, My Collections, and Solutions and My Space.
- Users with All Project Assistants or Frontend roles, including inherited wildcard domain roles, can access organization-level solution listings. Additionally, custom roles can now restrict visibility to specific solutions, ensuring users only see solutions explicitly assigned to them.
- Conversations linked to solutions that users no longer have access to remain readable, but execution actions are disabled.
- Improved access awareness by hiding the My Creations card when you only have a frontend role.
- Improved status labels in Build so “Published/Unpublished” read as “Saved/Not Saved,” keeping the existing logic and avoiding confusion about what’s published and what’s not.
- New Console entry in the Manage menu so you can jump to the Console for your environment regardless of role.
- New My Creations coverage so you can view and run Console Assistants and RAG Assistants inside the Run section of the Agentic Hub.
- Improved My Creations execution so you can filter and run Agents, Flows, Assistants, and RAG consistently with unified type mapping and run validation.
- Other RUN Section Improvements:
- Improved conversation banners so you can see a clear message when a solution is hidden by role scope and the chat is read-only.
- Fixed text wrapping in chat so long lines and bullets no longer break mid-word or overflow.
- Flows
- New Flows publishing from Build with a guided 4‑step wizard so you can configure avatar, execution permissions, chat sharing, and taxonomies before publishing and making them available for the whole organization.
- Improved Solution Detail for Flows with a dedicated layout and “Run Flow” actions distinct from Agents.
- Improved Flows editor branding with the new Glob.AI OS logo inside the Build section.
- API
- New quota visibility for frontend-role in Access Control API users so you can retrieve remaining quotas at the organization and project levels via /active and /latest, including access through domain wildcard.
- Improved Terms & Conditions API behavior so frontend-role users can query status without forbidden errors; acceptance remains inapplicable for that role.
- New middleware proxy endpoint so you can publish Flows from Build while the request is forwarded to the API.
- LLMs
- New OpenAI model so you can use gpt-5.5 for multimodal chat with tools, streaming, and advanced reasoning.
- New Claude Opus 4.7 availability across Anthropic, Vertex AI, Azure AI Foundry, and AWS Bedrock (US/EU/global variants).
- Improved catalog hygiene by hiding grok-4-1-fast-non-reasoning and grok-4-20-non-reasoning in Azure AI Foundry.
- New Azure AI Foundry support so you can use Claude Opus 4.6 in enterprise environments.
- Fixed cost schemas so reported pricing is correct for:
- Vertex AI image models “Nano Banana” (gemini-2.5-flash-image) and “Nano Banana Pro” (gemini-3-pro-image-preview).
- Azure OpenAI gpt-5, gpt-5-mini, and gpt-5-nano families.
- Console
- Improved API Token descriptions by rejecting control characters and the characters <, >, . and the null byte, with clear validation messages.
- Fixed organization deletion so removal completes cleanly and you can recreate organizations with the same name.
- Improved organization settings labels so “Station/Lab/Workspace” concepts are replaced with clearer options like “Publish agents in the Gallery” and “Share conversations.”
- Workspace Legacy
- Improved header links and hub menus so that Workspace Legacy reflects the hubs available based on user permissions.
- Improved branding by updating the Workspace Legacy logo to the new Glob.AI OS assets.