v6.0.0

SEMOSS 6.0.0 Release Notes

Release date: November 1, 2026
Previous release: SEMOSS 5.4.0, released July 26, 2026.

SEMOSS 6.0.0 expands the platform for building agents, automating workflows, and working with enterprise data and communications. This release brings a richer workbench, topic-based collaboration, scheduled routines, model routing, and broader governance and audit controls.

Upgrading an existing installation requires preparation. H2 databases must be migrated before starting 6.0.0. The release also moves to Tomcat 11/Jakarta APIs and Python 3.14, removes legacy KServe/NER integrations, and changes parts of the security database schema.

Highlights

• Agents and the workbench: Streaming agent runs, subagents, human delegation, run analysis, custom tools, Git, terminal access, notebooks, browser automation, and Office document editing.
• Workflow Automation: A dedicated project type with visual authoring, loops, decision routing, parameterized database actions, durable run history, file outputs, and version rollback.
• Collaboration: Topic-linked chats, searchable email and Teams context, calendar and file integrations, memory, and built-in and custom routines.
• Model management: Model routers, catalog-driven imports, pricing and capability metadata, reasoning settings, provider built-in tools, and expanded provider support.
• Enterprise governance: Rule-based model limits, guardrails across engines, SQL query policies, bulk access management, enterprise usage dashboards, and searchable audit trails.
• Runtime modernization: H2 2.4.240, Tomcat 11, Java 25 container images, Python 3.14, React 19, and updated dependencies.

Breaking changes and upgrade requirements

1. H2 1.4 databases require migration before startup

SEMOSS upgrades H2 1.4.200 to H2 2.4.240. H2 1.4 database files cannot be opened directly by the new runtime. Replacing the H2 JAR or enabling a compatibility mode does not convert an old database. Migration requires exporting with the old driver and importing with the new driver. See the H2 migration documentation.

Use the SEMOSS H2 Migration Service before starting the upgraded deployment. Follow the local upgrade runbook for filesystem installations or the Kubernetes cloud migration guide for remote storage.

Which data is affected

Database or storage location	Upgrade action
H2-backed system databases and user database engines under db/	Migrate the existing H2 files.
Engine audit databases	Include them in the db/ migration pass.
Project insight metadata and persisted saved-insight frames, including H2_Store_*.mv.db files under project/	Include them in the separate project/ migration pass.
H2 files in supported S3, Azure Blob, GCS, or MinIO storage layouts	Migrate the remote copies before upgraded nodes can pull them.
InsightCache files	The utility skips these rebuildable caches. Persisted saved-insight frames remain migration targets.
Custom application H2 files outside recognized layouts	Review separately; the utility reports unrecognized targets for manual review.
PostgreSQL or other non-H2 databases	No H2 file conversion is needed for those databases. Still inventory retained H2 files, audit stores, and project/frame stores.

Required upgrade sequence

1. Plan a maintenance window and stop every SEMOSS/Tomcat instance that can access the affected files or cloud storage. Keep them stopped for the real migration; do not run old and new nodes against the same data.
2. Back up the deployment and databases. Keep a full rollback copy of db/ and project/, configuration, and any external system databases. Store migration backups outside the SEMOSS home. For cloud deployments, retain a recoverable copy of the affected storage objects.
3. Run and review dry runs. Local installations require separate passes against the actual db/ and project/ directories, each with its own backup directory. Review every failed or unrecognized target before the real run.
4. Run the migration and save its reports. Run as the OS user that owns the SEMOSS files. Follow the runbook's disk-space guidance: local work storage needs about three times the H2 data size, and the source filesystem needs room for at least the largest database's temporary import.
5. Verify the converted data and deploy 6.0.0. Review the real-run reports and rerun the prescribed verification. For exploded Monolith deployments, remove the old h2-1.4.200.jar; the deployment must contain h2-2.4.240.jar without the old driver alongside it.
6. Start and smoke-test the deployment. Verify login, system databases, a user database engine, an existing insight, and a saved insight with an H2 frame. Check logs for 90048/unsupported file-version and 90113/unsupported connection-setting errors.

Rollback requires the old data as well as the old application. H2 1.4 cannot open converted 2.x files. Stop SEMOSS, restore the pre-upgrade files and any system databases changed by startup, and restore the previous application/configuration. Keep the rollback copy and migration backups until the upgrade has been accepted.

H2 SQL and connection compatibility

• H2 2.x reserves additional SQL keywords, including VALUE, KEY, YEAR, and USER. SEMOSS updates its reserved-word handling; custom SQL and schemas must quote identifiers where necessary.
• Numeric and Boolean values are no longer interchangeable in regular mode. Use Boolean literals such as TRUE and FALSE instead of comparisons to 1 and 0.
• Custom queries against INFORMATION_SCHEMA, identity/sequence syntax, and assumptions about data types or casts may need changes. Review the H2 SQL migration guidance.
• H2 1.x URL settings such as EARLY_FILTER, LOG, UNDO_LOG, MVCC, and MULTI_THREADED are unsupported. SEMOSS strips known obsolete settings at connection time, and the migration utility updates supported H2 profiles. Review custom .smss generators and deployment scripts so they do not reintroduce them. This cleanup does not replace database-file migration.
• Deployments using an external H2 TCP server must coordinate its version with the upgraded clients.

2. Tomcat 11, Jakarta APIs, and Java

The application moves from Tomcat 9 to Tomcat 11. Java EE servlet and REST imports move from javax.servlet.*/javax.ws.rs.* to jakarta.servlet.*/jakarta.ws.rs.*.

• Deploy the new Monolith WAR to a compatible Tomcat 11 installation; an existing Tomcat 9 installation is not a compatible target.
• Rebuild custom filters, servlets, authentication extensions, and Java integrations against the current SEMOSS and Jakarta dependencies. Merge custom web.xml, filter-chain, SSO, and container settings into the updated deployment configuration.
• Official container builds move from Java 21 to Java 25. Semoss and Monolith Maven compiler settings continue to target Java 21 bytecode; the container JDK change is distinct from the source target.

See the Tomcat migration guides for container-specific changes.

3. Python runtime and custom environments

The SEMOSS Python package now requires Python 3.14 or newer, increased from Python 3.12. Container Python builders also move to 3.14.

Rebuild custom Python environments using the 6.0.0 dependency definitions and validate application-specific packages, native wheels, model clients, and custom tools. Reusing an older Python environment does not satisfy the new package requirement.

4. Legacy KServe and NER integrations removed

The legacy KServe/ZooKeeper model-serving implementation and its UI are removed. Removed engine types include KSERVE_VISION, KSERVE_IMAGE_EMBED, KSERVE_IMAGE, KSERVE_TTS, and NER. Associated legacy deployment/status reactors, the NER reactor, the Python model wrapper's .ner() method, and the model.ner automation node are also removed.

Replace affected engine definitions, Pixel calls, Python integrations, and workflow nodes with supported model-engine integrations before upgrading. Existing custom Java extensions that depend on the removed classes must be updated and rebuilt. See Semoss #3213.

5. Security database schema and access behavior

Favorites and visibility are now user preferences. They are stored in USERENGINEPREFERENCE, USERPROJECTPREFERENCE, and USERINSIGHTPREFERENCE, keyed by user and login type, rather than permission rows. Favoriting an item no longer creates an access grant. Startup removes the legacy preference columns and permission rows that have no permission level. Update custom reports or integrations that read those columns; users may need to reapply legacy favorites and hidden-item preferences. See Semoss #3144.

Model limits use a new rule-based schema. Startup migrates legacy per-user and per-engine limit columns into MODELUSAGELIMIT and removes the old columns after the copy. Update scripts and reports that depend on the former USAGERESTRICTION, USAGEFREQUENCY, MAXTOKENS, MAXRESPONSETIME, or corresponding user-level columns. Review migrated limits and billing-source assignments through the new administration views. See Semoss #3162.

These startup schema changes also affect installations whose security database is hosted in PostgreSQL or another external database. Back up that database before the first 6.0.0 startup.

6. Frontend, SDK, and MCP integrations

• React 19: Rebuild custom frontend packages and check compatibility of React integrations. SemossWeb builds require Node.js >=24.4.0 <25 and pnpm 10.13.x.
• MCP metadata: Internal and remote MCP metadata no longer emit SMSS_PROJECT_ID and SMSS_PROJECT_NAME. Update custom consumers to use SMSS_ENGINE_ID and SMSS_ENGINE_NAME. See Semoss #2829.
• SDK room and agent APIs: The release adds RoomStore, AgentStore, and shared login state. Integrations built against intermediate 6.0.0 snapshots using the earlier function-based room/agent helpers should adopt the current exports and types.
• Media inputs: RunAgent uses media as the canonical file/media input key. The legacy image key remains an alias. Use media for new integrations and files beyond images.

New features and improvements

Agents, Playground, and MCP tools

• Separate Chat and Agent execution modes, with saved-agent selection independent of the execution-mode switch.
• Streaming agent progress and tool results, inline tool grouping, run refresh, activity views, parent/subagent relationships, and administrator access to agent-run records.
• Agent effectiveness metrics and model-assisted run assessments for inspecting execution quality.
• Subagents and human delegation, including assignee lookup and delegated-request context in the receiving room.
• Deferred tool loading, provider built-in tools, selective disabling of default tools, and room/workbench-specific toolboxes.
• Custom workbench tools, remote MCP configuration, richer tool-schema and parameter editors, and clearer tool titles and resolution errors.
• File and multimodal tool results, including file references and native tool views for mail, calendar, memory, and other supported components.
• Configurable context compaction, more reliable model/tool cancellation, and improved interruption handling.
• Message-content search with pagination, conversation downloads, tool-aware suggested questions, and clearer reasoning, cache, and context-usage information.

Workbench, notebooks, and application development

• A shared workbench brings together code, files, assistants, SQL, and resource panels, with terminal access available across workbenches.
• Project Git integration, commit history, and commits when reverting application changes.
• A Notebook project type and catalog, plus interactive Jupyter .ipynb viewing and editing, code/Markdown/raw cells, outputs, and notebook downloads.
• Chat code blocks can be run, saved as files, or added to a notebook. Notebook agents can find and update notebooks in project assets.
• Browser automation adds multi-tab sessions, selected-text/context capture, AI-assisted actions, saved recordings, and improved session cleanup.
• In-browser PowerPoint viewing and editing, presentation-building/review agents, and expanded Office document viewers and editing controls.
• Project templates, template administration, improved agent creation and shared-definition editing, and additional platform agents and skills for app building, data, documents, and tools.
• Build/publish improvements distribute custom reactors and portals across containers, preserve application commit history, and improve asset copying and project import/export.

Workflow Automation

• Workflow Automation is a dedicated project type with visual node and scope authoring.
• Loop nodes, configurable JEV decision routing, nested-variable completion, clearer run inputs, and notification configuration.
• Direct Java execution with improved scope transport and more reliable loop execution.
• Durable run history, run implementation snapshots, file outputs, and structured storage outputs.
• Database actions support query parameters and frame-backed outputs.
• Version listing, version-history panels, and rollback from the automation editor.

Collaboration and routines

• A Collaboration experience connects work across topic-linked chats, email, calendar, Teams conversations, and files.
• Guided onboarding imports and organizes source conversations into topics and people, with topic review, briefs, and background checks. Onboarding can continue past individual threads that fail sorting or filing.
• Topic-linked chats, shared threads, memory, global search, and source context help the assistant work with the relevant conversation history.
• Microsoft 365 integration covers Outlook mail and folders, whole email threads, attachments, calendar events, Teams threads, and OneDrive/Teams files.
• A Today view and topic work areas support daily work discovery. Drafted email can open in the editor, and owners can choose the Outlook folder used for drafts.
• Built-in routines include daily briefs, weekly meeting agendas, mail-triggered expense review against guideline documents, and event planning.
• Custom routines support schedules and time zones, instructions and preferences, attached documents, skills, and direct results.
• Routine history and live run views show execution progress and failure details.
• Connector settings, popup sign-in, refreshed login state, and calendar-source handling are improved.

Models and inference

• Model routers distribute requests across existing model engines, with clearer weighted routing and JEV/TypeSafe decision-based routing.
• Catalog-driven model imports, matching and resetting metadata on existing engines, improved model-ID suggestions, and a TypeSafe model catalog.
• Model metadata includes capabilities, modalities, serving/model providers, pricing, reasoning settings, and provider built-in tool configuration; metadata is included in import/export flows.
• Expanded NVIDIA NIM support and Vertex AI batch inference for Claude and Gemini staged through GCS.
• OpenAI-compatible Responses support in the model helper and HTTP filter, decisions-endpoint support, and authenticated in-platform Python access to SEMOSS-hosted model endpoints.
• Better enforcement of model input modalities, configurable Claude cache TTL, and reporting of cache-read, cache-write, and reasoning tokens.
• Improved provider/tool compatibility, including Amazon Nova tool-call sequencing, Gemini file deduplication, and uncompressed TypeSafe requests by default.

Connectors, databases, and storage

• New Bing and Brave search function engines and a ServiceNow function engine.
• Microsoft mail and calendar function engines, Exchange through Microsoft Graph, directory lookup, and Graph change-notification subscriptions/webhooks.
• Teams and SharePoint storage engines, plus Microsoft 365 and Google Workspace connector selection in Playground.
• Selected cloud storage engines and central cloud storage move from rclone-based operations to native provider SDKs, with compatibility fixes for existing storage credentials and container operations.
• PostgreSQL/pgvector engine support and improved embedding insertion.
• Parameterized SQL execution, shared JDBC parameter binding and resource cleanup, and faster schema synchronization.
• SQL scripts can execute as one transaction, with SQL-write auditing and configurable query-policy rules in the workbench.
• Updated database drivers and fixes for database-type normalization and failed-project cleanup.

Permissions, teams, governance, and audit

• Bulk access management, validated permission batches, effective-permission views, and catalog filtering by permission and creator.
• Team managers, My Teams, owner-approved team access, member previews, and copying people between team lists.
• People search can use existing SEMOSS users or the Microsoft directory, with provider-name normalization and improved group access queries.
• Separate administration controls for creating project types, including agents and skills.
• Model usage limits support rule/source definitions and reporting of cost and the source billed for each call. GetUserModelUsage includes COST_MICROS.
• Guardrails across engine types, policy and on-topic guardrails, self-harm/aggression checks, SQL guardrails, and JEV policy import.
• Guardrail hit limits, locked-user handling, hit auditing, configurable tool-continuation behavior, and judge-error verdict handling.
• Enterprise usage dashboards, searchable activity logs, agent-run inspection, and named audit events for engine/project exports and other user actions.

Catalogs, sharing, and user experience

• Shared engine/project overview components, refreshed catalog artwork, themed avatars, and image uploads.
• Project and engine rename actions, clearer display names, consolidated settings actions, and a Share App dialog.
• Expanded theme configuration, including dark-mode logos/icons and form-driven Playground settings.
• Improved file-explorer actions, workbench panel behavior, mobile room settings, chat loading, and code-block controls.
• Users can remove their default model selections from My Profile.

Reliability and security fixes

• Agent-run locking and Redis refresh coordination reduce conflicting updates during streaming; room projections now expire and are evicted when sessions end.
• Cancellation propagates through model streams, pending tools, and agent work more reliably; failed cancellation and tool errors are surfaced more clearly.
• System-database writes use transactions, and failed project creation rolls back more consistently.
• Engine-secret changes close/reload engines across nodes; engine cloning/restart and cluster-wide project Python reload are available.
• Project portals and custom reactors are released to every container more reliably.
• Python guardrails terminate their Python processes when closed. Rendered Python figures can be returned inline without GUI backends terminating the process.
• Request-controlled file paths and headers are handled more safely; regex-injection, SQL/TLS, and content-type handling are hardened.
• Stored credentials use updated hashing, with legacy credentials migrated after a successful login.
• SDK requests preserve useful Pixel error messages, improve CSRF/session handling, and support an opt-in cookie jar for headless use.
• Fixes cover PostgreSQL Playground search, notebook refresh/editing, file uploads and downloads, OneDrive traversal, shared-app routing, and Office/Visio save compatibility.
• Java, Python, frontend, and container dependencies receive security and compatibility updates, including database drivers, cryptography, networking, document processing, and model clients.

Deployment and build changes

• Container builders use Tomcat 11, Java 25, and Python 3.14, with refreshed Node and Maven tooling.
• The all-images build dispatcher targets UBI9 and Ubuntu 22.04, 24.04, and 26.04. UBI8 and Amazon Linux 2023 were removed from that aggregate dispatcher; deployments using those variants should review their image-build path.
• FIPS-oriented builds add FIPS 140-3 provider configuration and separate FIPS dependency packaging. Pair a FIPS container base with the corresponding FIPS-profile WAR; validate custom cryptography and SSH integrations against that build.
• Node execution is available to agents, with updated sandbox-path, process-limit, startup dependency, and Kubernetes environment handling.
• Python-builder jobs refresh weekly and use current uv tooling so Python patch versions are not held back by an old builder tool.
• Merge updated deployment templates and feature configuration for collaboration, Microsoft Graph notifications, model usage controls, guardrails, and Redis room expiration where those features are enabled.

Full change history

These notes cover committed changes after v5.4.0 in Semoss, Monolith, and SemossWeb, reviewed against upstream dev on November 1, 2026. The links below pin the reviewed commits so the release scope remains reproducible:

• Semoss: all changes since v5.4.0
• Monolith: all changes since v5.4.0
• SemossWeb: all changes since v5.4.0

Deployment changes were also checked against semoss-artifacts. The H2 upgrade instructions refer to the separate H2 Migration Service.

What's Changed

• fix: capitaliztation in pipelines by @kunal0137 in #2811
• Add self-harm/aggression guardrail by @jareallen213 in #2796
• Remove codex from Dockerfiles by @ryanweiler92 in #2813
• fix: stop GUI backends killing the python process and return rendered figures as inline images by @themaherkhalil in #2816
• Prevent compaction on invalid message states by @tlokeshrao in #2803
• Agent page updates by @tevanburen in #2789
• fix: resolving issue with github action not assigning var for download url by @themaherkhalil in #2819
• feat: update tomcat from 9.0.119 to 9.0.200 by @themaherkhalil in #2820
• Playwright socket: room level MCP toolboxes, platform system app, multi-tab and selected text capture by @ibrahimmahaba in #2779
• Add model capability metadata to the security database by @ryanweiler92 in #2818
• chore: updating image to 6.0.0 snapshot by @themaherkhalil in #2826
• feat: adding remote mcp details and fixing remote tool execution by @themaherkhalil in #2827
• Feat/playwright system app recording save by @ibrahimmahaba in #2825
• chore: updating image tag by @themaherkhalil in #2828
• chore: standardize to engine instead of project for mcp meta keys by @themaherkhalil in #2829
• chore: app rename from _ to - by @themaherkhalil in #2830
• Updating Py Packages 08-03-26 by @ryanweiler92 in #2831
• feat: update clickhouse jar, append additional props to general method, update log statements by @themaherkhalil in #2832
• fix: loading projects w/o pulling from cloud for platform projects, remove deprecated keys from platform project smss files by @themaherkhalil in #2834
• feat: updating from javax to jakarta by @themaherkhalil in #2835
• feat: update tomcat9 to tomcat11 by @themaherkhalil in #2837
• fix: add meta directory by @themaherkhalil in #2839
• Ignore functions with leading underscore when making python MCP by @ckelly17 in #2836
• Surface cache read/write and reasoning token usage on ResponseMessage by @ryanweiler92 in #2838
• feat: update java21 to java25 in all images by @themaherkhalil in #2840
• chore: rename workflow name by @themaherkhalil in #2841
• chore: rename action by @themaherkhalil in #2842
• fix: ensure the user asset folder exists, remove legacy project restructure which broke out database and project insights by @themaherkhalil in #2843
• chore: remove legacy json processor code by @themaherkhalil in #2844
• chore: remove legacy playwright code by @themaherkhalil in #2845
• feat: adding proxy on all database engines by @themaherkhalil in #2846
• Store parent run IDs for RunAgent subagents by @ryanweiler92 in #2848
• Add On-Topic Guardrail by @jareallen213 in #2784
• Output Policy Compliance guardrail by @manamittal in #2817
• Fix stray javax.ws.rs import to jakarta.ws.rs by @tevanburen in #2849
• Agent activity log, subagent runs, and automatic room naming by @ryanweiler92 in #2851
• fix: return message history on run agent wait by @kunal0137 in #2847
• Browser automation ai and extract context enhancements by @ibrahimmahaba in #2850
• feat(vector): add pgvector engine support and fix embedding insertion by @FarhanOhe03 in #2824
• chore: moving cron schedule by @themaherkhalil in #2854
• fix(agent-clone): preserve workspace config & agent data when cloning an agent by @FarhanOhe03 in #2852
• feat: using db type for proper query reactor, always attempting to send title optional value, reducing filename metadata path by @themaherkhalil in #2855
• Updating Model Settings Handling by @ryanweiler92 in #2857
• feat: changes to support fips - pom modifications + image building changes by @themaherkhalil in #2858
• fix: check TOMCAT_HOME properly set for setenv.sh by @themaherkhalil in #2859
• fix: separate java.security.fips so build process using maven can pull using standard java.security by @themaherkhalil in #2860
• fix: prevent app cloning when read only by @tlokeshrao in #2856
• Model Metadata Mapping by @ryanweiler92 in #2861
• Manual syncing of model metadata by @ryanweiler92 in #2863
• fix: switching to fips compliant hash for stored creds, migration happens after first login with legacy creds by @themaherkhalil in #2864
• Browser automation click save tabs by @ibrahimmahaba in #2862
• Updating Py Packages 08-10-26 by @ryanweiler92 in #2865
• Feat cancel llm stream kp by @tlokeshrao in #2785
• feat: create generic ask room to mirror ask playground by @kunal0137 in #2866
• fix: drop selector for hasportal and portalname by @kunal0137 in #2868
• fix: metadata not breaking load of engine by @themaherkhalil in #2871
• Provide closest matching model ids on model engine creation by @ryanweiler92 in #2870
• feat: adding in service now function engine by @themaherkhalil in #2872
• Preserve model metadata when exporting and importing model engines by @ryanweiler92 in #2873
• Reflection error on pooled db by @agmontem-devel in #2874
• Canonical agent run streaming, harness auto-compaction, and platform tool cleanup by @ryanweiler92 in #2867
• fix: add serializer for sqlexception in proxied invocation logging by @agmontem-devel in #2877
• Provider Built-in Tools: catalog, reactor, storage, and runtime plumbing by @ryanweiler92 in #2881
• feat(agent): add secure agent run audit support by @kunal0137 in #2882
• feat: adding the ability to create a notebook by @neelneelneel in #2883
• feat: adding default skills for mcp, allowing overriding for custom pixel_mcp.json by @themaherkhalil in #2884
• Fix OpenAi Image Gen spec and close engine on setting update by @ryanweiler92 in #2887
• fix: acquire the room lock when refreshing from redis to not override executions mid stream by @themaherkhalil in #2889
• Pass Thinking/Reasoning from model meta by @ryanweiler92 in #2890
• Drop Temperature model parameter when not supported by @ryanweiler92 in #2891
• fix: updating edit workspace for additional parameters by @kunal0137 in #2888
• Amazon Nova tool calling fails with "Model produced invalid sequence as part of ToolUse" by @ryanweiler92 in #2892
• feat: adding explicit project templates by @kunal0137 in #2886
• fix: not loading and locking on a room for the audit logs filter to a room by @themaherkhalil in #2893
• fix: update java fips security for bcfips by @themaherkhalil in #2894
• feat: adding in javascript and updated overview tab by @themaherkhalil in #2895
• Model Router engine — route requests across existing model engines by @JoshuaDivineBlessing in #2869
• Model router bootstrap by @ryanweiler92 in #2897
• fix: agent streaming by @tevanburen in #2880
• fix: correcting reactor name by @themaherkhalil in #2898
• Fix stale system prompt shown in Agent UI for system agents by @ryanweiler92 in #2900
• Edit Guardrails Configuration Reactors by @ryanweiler92 in #2902
• Add model pricing metadata to the security database by @ryanweiler92 in #2903
• fix(agent): respect named subagent default model by @kunal0137 in #2905
• feat: updated skill files, update image= to media= for models, project meta values filtered by type, copy asset allows override parameter by @themaherkhalil in #2904
• fix: agent seeder will also remove for changes in default agent by @themaherkhalil in #2907
• fix: adding in python skill for getting ROOT variable, fix legacy image -> media in engine usage by @themaherkhalil in #2908
• feat: engine workbench room mcps by @ppatel9703 in #2899
• Built In Tools Doc by @ryanweiler92 in #2909
• chore: using method for taking token details to map by @themaherkhalil in #2911
• feat: adding more tools to publish project, adding preference for normal save over baes64 version by @themaherkhalil in #2913
• fix: accounting for azure models by @themaherkhalil in #2914
• fix: app bootstrap skill by @themaherkhalil in #2917
• Changing RunAgent param images to media by @ryanweiler92 in #2918
• fix: named subagents should not expose a changed system prompt by @kunal0137 in #2919
• Moving selected engines from RunAgent flow to System Prompt by @ryanweiler92 in #2920
• feat: move the storage engines and CentralCloudStorage from rclone to native cloud SDKs by @ppatel9703 in #2628
• chore: proper file headers on files by @themaherkhalil in #2921
• fix: markdown syntax error in storage engine usage by @themaherkhalil in #2922
• feat: adding default json for notebook by @themaherkhalil in #2923
• fix: switch to previous s3 keys, google works with gcs or gcp prefixed keys by @themaherkhalil in #2924
• feat: adding version for azure storage, fix: allowing copy of entire container by @themaherkhalil in #2925
• feat: interact with files ms teams by @themaherkhalil in #2926
• feat: adding teams and sharepoint storage engines by @themaherkhalil in #2927
• fix: dont break on container missing, only remove empty dir on sync storage to local by @themaherkhalil in #2929
• Allow deployment-level exemptions from ADMIN_ONLY_PROJECT_ADD by project type by @Kaveh-Vakili in #2807
• Resolve display title for runagent tool items by @tevanburen in #2928
• fix: setting boolean properly so once function exists no longer rechecks, switching from sync to lock by @themaherkhalil in #2931
• Apply model catalog metadata to existing engines (match + reset to defaults) by @ryanweiler92 in #2932
• feat: bing and brave web search function engines, ability to define custom MCP details for ExecuteFunctionEngine reactor at engine level by @themaherkhalil in #2933
• RunAgent dispatches provider-executed builtin tools (web_search) as local MCP tools by @ryanweiler92 in #2934
• Refactor IS_TEMPLATE update to use PreparedStatement by @kunal0137 in #2937
• Catalog-driven model import: ListStaticModelCatalog pixel + serving-host pricing for Vertex/Bedrock/Azure by @ryanweiler92 in #2938
• feat: allow defining the cache control ttl for claude by @themaherkhalil in #2939
• Add media tool response by @ckelly17 in #2910
• fix: enforce configured model input modalities by @tlokeshrao in #2896
• Guardrail fixes enhancements by @manamittal in #2879
• Compilation fix modality enforcer by @ryanweiler92 in #2940
• Python upgrades: openai 3.5.0 (HTTPX2), anthropic 1.1.0 + TLS compatibility fixes by @ryanweiler92 in #2941
• fix: proper sslcontext reference for httpx and pips replacement by @themaherkhalil in #2944
• Python Package Updates by @ryanweiler92 in #2948
• Search Bar Enhancement: add message content search backend by @Kaveh-Vakili in #2764
• feat: disable specific default tools by @kunal0137 in #2947
• feat: mail function engines by @themaherkhalil in #2950
• feat: adding exchange via graph api by @themaherkhalil in #2952
• chore: removing unused classes by @themaherkhalil in #2953
• chore: remove unused code, try-with-resource, log fixes by @themaherkhalil in #2954
• feat: exposing microsoft outlook functions, adding tracking of mail outside of smtp protocol by @themaherkhalil in #2955
• feat: breakout mail logic into independent classes by @themaherkhalil in #2956
• chore: cleanup create function engine calls by @themaherkhalil in #2957
• chore: rework package structure for helper mail classes by @themaherkhalil in #2958
• fix: proper method signature of askRoom, guardrail pipeline fixes by @themaherkhalil in #2961
• Add in-platform ServerClient stand-in to the ai_server shim by @ryanweiler92 in #2962
• Agent run effectiveness metrics + LLM-as-judge assessment by @ryanweiler92 in #2943
• feat: adding microsoft calendar reactors, method to create room pixel mcp tools by @themaherkhalil in #2964
• feat: adding sql guardrail, handle multi sql by @themaherkhalil in #2966
• Add omega-memory dependency to pyproject.toml by @kunal0137 in #2969
• Project templates by @neelneelneel in #2959
• feat: update jsqlparser jar and parsing logic by @themaherkhalil in #2972
• fix: react template package updates by @kunal0137 in #2976
• fix: postgres playground search and performance improvement by @kunal0137 in #2977
• feat: add zk engine back by @themaherkhalil in #2978
• Updating Model Metadata for 09/04/2026 by @ryanweiler92 in #2981
• fix: removing insight from rest and streaming rest, adding correct subtype by @themaherkhalil in #2985
• feat: initial project git by @neelneelneel in #2965
• fix: deduplicate files when building for gemini by @ckelly17 in #2979
• fix: pretty printing off on index.html rewrite so agent can string replace, updated skills and harness instructions by @themaherkhalil in #2986
• Updating Model Metadata 09-08-26 by @ryanweiler92 in #2988
• Updating Python Dependencies 09-08-26 by @ryanweiler92 in #2989
• feat: allowing skills with multiple files to be discovered, improving instructions for app building by @themaherkhalil in #2991
• fix: consolidate agent run room locking into a single registry by @themaherkhalil in #2993
• Agent Node.js Execution Environment by @ryanweiler92 in #2960
• Update react template by @neelneelneel in #2994
• Update react template by @neelneelneel in #2995
• Set var for enabling node tools in kube env by @ryanweiler92 in #2996
• Allow file inputs if attachment column is true by @ryanweiler92 in #2997
• fix: proper logic and comparisons to pull / republish project portals across containers by @themaherkhalil in #2998
• Copy assets anywhere by @ryanweiler92 in #2999
• Fix/snyk regex injection by @ibrahimmahaba in #2987
• fix: remove dead rpa code and rework the scheduled pixel job by @ibrahimmahaba in #2992
• Fix/snyk dependency upgrades by @ibrahimmahaba in #2967
• fix: upgrade remaining vulnerable dependencies (netty, parquet, opentelemetry, log4j) by @themaherkhalil in #3002
• chore: proper file headers, minor code formatting/edits by @themaherkhalil in #3003
• feat: update fips jar and docker image by @themaherkhalil in #3004
• fix: readd original text for legacy to use in displaying the sql stat… by @antgibson96 in #3001
• chore: Updating cron job times by @tymayer in #2983
• feat: enable openai in platform python process using authenticated user credentials by @themaherkhalil in #3005
• fix: adding if process is engine to allow bypassing of overriding modules, proper ulimit syntax and changing to ULIMIT_PROCESS by @themaherkhalil in #3006
• chore: adding action to dispatch and create all final images by @themaherkhalil in #3007
• fix(security): upgrade vulnerable Node agent dependencies by @ibrahimmahaba in #3008
• Optional Tool Continuations Skipping in Guardails by @manamittal in #2982
• Updating reserved params for OpenAI audio client by @ryanweiler92 in #3009
• Static auto greet by @manamittal in #2970
• Updating Py Packages 09-16-26 by @ryanweiler92 in #3012
• fix: honor agent run cancellation - check interrupt flag and clear waiting for payload struct response epoc by @themaherkhalil in #3013
• feat: set SANDBOX_PATH by @kunal0137 in #3014
• fix(deps): upgrade MariaDB and Snowflake JDBC drivers by @ibrahimmahaba in #3010
• feat: add space as run agent parameter for projects or user by @kunal0137 in #2963
• Reduce Node friction for agent use by @ryanweiler92 in #3011
• Adding pptx agent smss files by @ryanweiler92 in #3019
• feat(agent): add managed Python default tool by @kunal0137 in #3017
• Pptx editor by @ryanweiler92 in #3022
• feat: workflow automation project type by @ppatel9703 in #2812
• feat: improve python automation skill, project seeder to update default skills/agent, clear legacy keys by @themaherkhalil in #3023
• feat: buildout of microsoft reactors by @themaherkhalil in #3025
• Commit history when building apps by @tlokeshrao in #3021
• Jev Support by @ryanweiler92 in #3026
• Engine Image Uploads by @ryanweiler92 in #3029
• feat: adding ability to subscribe to events for logged in microsoft, fix: use active token (auto renew if expired) by @themaherkhalil in #3031
• feat: exposing new agents and skills, ability for FE to append or override the agent system prompt with a workbench (room) prompt by @themaherkhalil in #3032
• Update PPTX Editing flow by @ryanweiler92 in #3033
• Add Frontend Design Platform Skill by @ryanweiler92 in #3034
• Update model.json 09-23-26 by @ryanweiler92 in #3036
• fix: adding buffer for max subscription times by @themaherkhalil in #3037
• feat: subagent human delegation by @kunal0137 in #3038
• feat(collaboration): resolve delegation assignee by name and add Find… by @kunal0137 in #3039
• fix: adding more projections in user search query, code formatting in other files by @themaherkhalil in #3042
• feat(automation): improve scope authoring and JEV routing by @ppatel9703 in #3035
• feat(collaboration): put the delegated request in the assignee room's… by @kunal0137 in #3040
• Upgrade BouncyCastle and freemarker dependencies by @ibrahimmahaba in #3030
• feat: notification system by @kunal0137 in #2721
• PPTX Updates to handle unterminated string error by @ryanweiler92 in #3047
• chore: switching logs to debug by @themaherkhalil in #3048
• fix: adding a proper error message when a login type is required by @themaherkhalil in #3049
• chore: fix license headers by @themaherkhalil in #3050
• feat(automation): add structured storage outputs and run files by @ppatel9703 in #3045
• fix: adding missing function to clear app imports from smssutil by @themaherkhalil in #3054
• chore: adding documentation by @themaherkhalil in #3055
• fix: properly supporting export/upload of agent, methods for checking if db exists by @themaherkhalil in #3056
• Multi modal tool response returns file reference by @ryanweiler92 in #3059
• feat: handle text files properly in media, update connector metadata on outlook, scope checks by @themaherkhalil in #3061
• feat(automation): fix insight issue over char limit by @ppatel9703 in #3063
• fix(security): validate header fields and restrict pipeline reactors by @ibrahimelnemr in #3015
• fix: resolve medium Snyk SQL and TLS findings by @ibrahimelnemr in #3018
• feat: return display names and descriptions from GetWorkspace by @themaherkhalil in #3064
• feat: adding reactor to get agent platform tools and hooks by @themaherkhalil in #3065
• feat: merge Database Maker backend updates into mcp_driver.py by @JoshuaDivineBlessing in #3051
• Update model catalog & Py Packages by @ryanweiler92 in #3067
• feat: filter project and engine lists by effective permission and creator by @themaherkhalil in #3068
• feat: catalog image refresh by @themaherkhalil in #3072
• feat: add enterprise usage reactors, filters, and audited exports by @themaherkhalil in #3073
• refactor(jdbc): centralize execution and parameter binding by @themaherkhalil in #3076
• fix(auth): normalize provider aliases and correct group access queries by @themaherkhalil in #3077
• refactor(jdbc): consolidate repetitive statement execution by @themaherkhalil in #3078
• feat: collaboration by @kunal0137 in #3058
• feat: attach files to email from collaboration by @kunal0137 in #3079
• refactor(automation): organize runtime and execution services by @ppatel9703 in #3075
• feat: collaboration summaries by @ryanweiler92 in #3081
• chore(docker): add unoserver and node-builder examples, enable node env by @themaherkhalil in #3084
• feat: collaboration system skills by @kunal0137 in #3085
• feat (collaboration): adding in date time for semoss status + jev cla… by @kunal0137 in #3086
• feat: deferred agent tools by @kunal0137 in #3087
• Fix: GetInsightActiveRoomModel in agent run by @ckelly17 in #3082
• feat: interact with brain from agent by @kunal0137 in #3095
• fix: upgrade mvn libs and maven in docker by @kunal0137 in #3097
• feat: add parameterized sql execution and shared jdbc cleanup by @themaherkhalil in #3098
• Collaboration Memory by @ryanweiler92 in #3099
• Collaboration isAssistantRoom by @ryanweiler92 in #3100
• Feat/automation loop node by @ppatel9703 in #3080
• fix: context window management by @kunal0137 in #3102
• feat(connectors): shared mail and calendar reactors with component views by @themaherkhalil in #3103
• fix(connectors): add the Microsoft reactors missing from #3103; share value helpers by @themaherkhalil in #3104
• fix: let the notebook analyst find and write notebooks under public by @themaherkhalil in #3105
• feat(auth): look up users in the Microsoft directory or the security database per request by @themaherkhalil in #3106
• fix: update tomcat version by @kunal0137 in #3107
• Collaboration: PPTX authoring, Pixabay images, and tool-call reliability by @ryanweiler92 in #3127
• feat(auth): team managers, owner-only team access, Graph-verified directory users by @themaherkhalil in #3128
• feat:jev guardrail by @kunal0137 in #3096
• feat(collaboration): chats linked to topics, topic brief, background topic check, onboarding review by @kunal0137 in #3129
• [#3113] feat(collaboration): staged onboarding topics, areas and a live import by @kunal0137 in #3131
• feat: add trace reason to the return for json logic by @travon-speller in #3134
• [#3113] fix(collaboration): widen BRAIN_TOPIC_PERSON.ROLE_LABEL to 255 by @kunal0137 in #3132
• [#3113] fix(collaboration): give the thread assistant the owner's profile in its system prompt by @kunal0137 in #3135
• feat(automation): execute database outputs through frames by @ppatel9703 in #3130
• feat(automation): support configurable Jev routing by @ppatel9703 in #3126
• fix: room naming by @kunal0137 in #3136
• fix: refresh sync improvementa by @kunal0137 in #3138
• Collaboration room thread mapping by @ryanweiler92 in #3140
• feat(auth): validate and save permission changes as permission batches by @themaherkhalil in #3141
• feat: add trim, blank fallback and lowercase key helpers to ValueUtils by @themaherkhalil in #3142
• chore: remove al2023 and ubi8 builds by @themaherkhalil in #3143
• feat: keep favorites and visibility in user preference tables by @themaherkhalil in #3144
• feat: let owners see team members first, copy between team lists by @themaherkhalil in #3145
• Collaboration Routines by @ryanweiler92 in #3147
• feat: rewrite Automation around direct Java scope by @ppatel9703 in #3148
• feat: add tool-aware playground suggestions by @ibrahimelnemr in #3151
• fix(automation): stabilize loop execution and scope transport by @ppatel9703 in #3153
• Make pptx agent reviewer optional by @ryanweiler92 in #3154
• fix(automation): expose run implementation snapshots by @ppatel9703 in #3155
• chore: upgrade tomcat version by @kunal0137 in #3159
• Collaboration: Updating Collab Routines by @ryanweiler92 in #3160
• Fix/collaboration global search by @ibrahimmahaba in #3158
• Updating NIM Support by @ryanweiler92 in #3161
• feat(model-usage): rework model usage limits into rules, sources and one guard by @themaherkhalil in #3162
• Audit trail backend by @ibrahimmahaba in #3152
• feat(usage): record whom each model call was billed to, report model cost by @themaherkhalil in #3164
• Feat collaboration draft folder by @kunal0137 in #3165
• Updating Py Packages 10-19-2026 by @ryanweiler92 in #3166
• Model router: send images to TypeSafe decision classifiers, keep agent turns on one model by @ryanweiler92 in #3169
• OpenAI Decisions support by @ryanweiler92 in #3170
• Vertex AI batch inference for Claude and Gemini, staged through GCS by @kunal0137 in #2984
• feat(collaboration): skill builder, direct routine results, calendar … by @kunal0137 in #3167
• feat: collaboration attachments on routines by @kunal0137 in #3171
• feat: SQL write auditing, one-transaction scripts and faster schema sync by @themaherkhalil in #3172
• feat: collaboration agent by @kunal0137 in #3173
• fix: release reactors and portals to every container by @themaherkhalil in #3175
• Collaboration: Adding drafts back to daily routine by @ryanweiler92 in #3176
• Fix: update container and jar dependencies by @kunal0137 in #3177
• feat(guardrail): JUDGE_ERROR_VERDICT and one BLOCK or PASS verdict for guardrails by @themaherkhalil in #3180
• Upgrade Python 3.14 by @ryanweiler92 in #3179
• Updating Python builders to 3.14 by @ryanweiler92 in #3181
• feat: Add /responses endpoint support to OpenAIResponsesHelper for Goose by @JoshuaDivineBlessing in #3178
• fix(model-usage): use the Redis counter store only when Redis is enabled by @kunal0137 in #3183
• fix(room): expire and evict Redis room projections by @kunal0137 in #3184
• fix: update node version by @kunal0137 in #3182
• build(deps): update x-pack-sql-jdbc, sharp, image-size and Zulu for c… by @kunal0137 in #3185
• feat(database-maker): tool views for MCP tools and app fixes by @neelneelneel in #3188
• Collaboration routines: reference documents for custom routines, and an inbox daily brief assembled by code by @ryanweiler92 in #3189
• build(deps): bump npm, jackson, amqp-client and undici for container … by @kunal0137 in #3193
• fix(model): send TypeSafe request bodies uncompressed by default by @kunal0137 in #3194
• fix: transactional writes for the system databases by @themaherkhalil in #3195
• fix: collaboration calendar source by @kunal0137 in #3196
• fix: add node to path by @kunal0137 in #3197
• Collaboration: Expense Review routine (mail-triggered, reviews approval requests against guideline documents) by @ryanweiler92 in #3211
• Removing old KServe implementation by @ryanweiler92 in #3213
• feat(engine): clone and restart engines, reload project python across the cluster by @themaherkhalil in #3214
• fix: python guardrails stop their python process when they close by @themaherkhalil in #3215
• fix: removing outdated test of methods that were removed by @themaherkhalil in #3217
• Collaboration: Expense Review Updates by @ryanweiler92 in #3216
• feat(audit): record engine and project exports by @themaherkhalil in #3223
• feat(agent): admin agent run reads, room close, and run login types by @themaherkhalil in #3225
• feat(automation): add version listing and rollback support by @ppatel9703 in #3200
• feat(guardrail): hit limits, user locks and hit auditing on every engine by @themaherkhalil in #3226
• Update pixabay system MCP to use project py by @ryanweiler92 in #3227
• fix(workspace): include display name in workspace details by @kunal0137 in #3228
• Blank Quartz's placeholder H2 admin/admin datasource credential by @terryozaki-wq in #3207
• Remove duplicate maven-compiler-plugin declaration from pom.xml by @terryozaki-wq in #3204
• Throw actionable errors for invalid RdbmsTypeEnum config values by @terryozaki-wq in #3205
• feat(collaboration): add event planning routine and routine run history by @kunal0137 in #3220
• H2 bump by @rithvik-doshi in #2332
• MCP-level resolution failures by @ckelly17 in #3190
• Return COST_MICROS from GetUserModelUsage by @ckelly17 in #3224
• fix(deps): bump vulnerable dependencies from image and Snyk scans by @kunal0137 in #3232
• fix: ci tomcat builder codebuild by @kunal0137 in #3233
• ci(docker): use latest uv in python-builder and rebuild weekly by @kunal0137 in #3234
• Fix/workspace legacy insights db by @kunal0137 in #3235
• feat: only INSIGHT projects keep an insights database by @themaherkhalil in #3236

New Contributors

• @Kaveh-Vakili made their first contribution in #2807
• @tymayer made their first contribution in #2983
• @terryozaki-wq made their first contribution in #3207

Full Changelog: v5.4.0...v6.0.0