Release Notes
HighByte Intelligence Hub Version 4.6 Beta
New Features:
Enhanced the MCP Server page to allow users to define and manage MCP Tool definitions. MCP Tools are now created on the MCP Server page with name, description, input parameters, etc. and linked to Pipelines for execution. Existing MCP Tools are migrated and continue to work after upgrade.
Added a new REST Data Server page to allow users to define and manage API endpoints. Endpoints can now be created on the REST Data Server page with URL, HTTP Verb, description, and request body schema, and linked to a pipeline for execution. Existing v1/ REST Data endpoints continue to work after upgrade.
All existing Pipelines with API Triggers are migrated to the new MCP Server and REST Data Server pages, and the API Trigger is removed from Pipelines. The Server pages support mass delete for cases where this migration is not needed (e.g., not using REST Data Server).
Added audit logging to the MCP Server to view when tools are called, by who, and with what arguments.
Added a new ‘agent_use’ action to control access to view and call MCP tools. Existing Pipeline permissions are migrated to the MCP Tools on upgrade.
Added a new Active/Active mode for High Availability (HA). In Active/Active both hubs execute Pipelines and host Server interfaces, but only the Primary publishes data to source systems. The Secondary buffers writes over a small window, and on switchover, flushes the buffer and continues as the Primary. This reduces the chances of data loss on failover.
Added the ability to exclude some parts of project configuration from synchronization in HA mode. This includes the project, variables, secrets, and pipeline and input state.
Added a new setting in HA configuration to control if a node is a preferred leader. In Active/Active mode, it’s common to have no preferred leader, avoiding unnecessary failovers.
Redesigned the Namespaces UI to allow for resizing and collapsing panels and making Namespace configuration and testing easier.
Added a new QuestDB Connection. The Connection uses the QWP protocol, and supports SQL inputs, Parquet file uploads, and the ability to write to and create tables.
Added a new Cognite Data Fusion Connection. The Connection supports both reading and writing Cognite Time Series and Assets. Outputs can be configured to write to existing nodes in Cognite or create Time Series and Assets as part of the write.
Added an AMQP 1.0 Connection for subscribing and publishing to brokers such as RabbitMQ via the AMQP v1.0.0 protocol.
Added support to Snowflake Streaming to use the new V2 High-Performance Architecture SDK. The SDK selection is an option on the existing Snowflake Streaming Connection.
Added a new History input to the Ignition Module Connection to support reading historical tag data with Ignition 8.3+ Gateways.
Added support for a new Historical Input to the OPC UA Client connection. The input can read raw or aggregated history over a chosen time range and processing interval.
Added OPC UA 1.04 support for complex data discovery. The Connection now discovers custom structures, enumerations, and unions through DataTypeDefinitions. The connection defaults to the legacy dictionaries discovery for servers that support both.
Updated Databricks Zerobus connections to version 1.6 and introduced ARM support.
Updated the BigQuery Connection to use the native JDBC driver provided by Google.
Enhanced OpenTelemery support to include audit logs when audit logs are enabled.
Added MCP Configuration tools for Namespaces, including publishing the Namespace configuration schema.
Added GitHub Applications Authentication support to Git backup.
Added a new JVM widget to the Dashboard to show machine statistics like memory, CPU, and threads. These statistics are also available as data in a new System.Internal.JVM reference.
Added a new Types interface to JavaScript expressions in the JavaScript stage, Model stage, and Instance expressions. This allows control of data types in expressions like Floats vs. Integers.
Added indexing to the Event Log, Audit Log, and Pipeline Activity SQLite files, improving query and truncation speed.
Removed the default buffer key configuration from new Timed and Size Buffer stages. Leaving the buffer key unset improves performance.
Added a new License name field to the license page.
Enhanced the SQL Connection inputs page to show the SQL query used by the input.
Added support for remembering the height of code editor fields used in JavaScript stages, Instance expressions, etc. for the browser session.
Fixes:
Fixed an issue where copying stages in a Pipeline lead to invalid sequences (e.g. Javascript_1, Javascript_2, etc).
Fixed an issue where int8 types in Amazon Redshift were incorrectly mapped to character(1) instead of smallint.
Fixed an issue where users could see entities they didn’t have permission to see on the dashboard.
Fixed an issue where server port configurations for the REST Configuration server, REST Data, MQTT, etc. could overlap, causing the application to not start.
Fixed an issue where an OPC UA connection could start publishing values from a different OPC UA connection's tags. This could occur if all the following were true: several connections used the same server and user identity, the server reissued a subscription ID that was in use before a restart, and the affected connection was slow enough that its connection check was delayed.
Fixed an issue where JSON Schema model imports would reject nullable arrays.
Fixed an issue causing Connection Input browses to not properly show a loading icon.
Fixed an issue where a Model stage with Expand Arrays enabled returned a single default-valued object when it received an empty array. The stage now outputs an empty array.
Fixed an issue where the Reference Panel did not organize Instances, Pipelines, and other elements into their Group As folders.
Fixed an issue where Model attributes marked Nullable that evaluated to null were dropped instead of being kept as typed null values. This applied to the Model stage and Instances. Non-nullable model attributes that evaluate to null now receive their default value.
Fixed an issue where the sort column and direction chosen on list pages were not remembered after navigating away and returning.
Breaking Changes:
Added a setting to the OpenAI Connector to choose between using the “/responses” and “/chat/completions” for backwards compatibility. Existing OpenAI connections that have a Base URL were migrated to the /chat/completions. Connections without a Base URL use /responses. If your custom LLM endpoint supports the Responses API, set API back to /responses.
The MySQL JDBC driver is no longer included with the Intelligence Hub due to licensing restrictions. MySQL connections, and JDBC Driver connections that use MySQL, will fail to connect after upgrading until the driver is downloaded, copied into the runtime/lib directory, and the runtime is restarted.
Project import, export, and sync now use a dedicated project claim. Users with the network:create claim have been automatically permissioned with project:import and project:export claims.
Security Patch Updates:
Frontend:
CVE-2026-101898: Defect that allows the HTTP/2 adapter to bypass the configured DNS lookup and proxy controls.
CVE-2026-101899: Defect that causes CIDR-form NO_PROXY entries to be ignored, enabling proxy exclusion bypass for internal IP ranges.
CVE-2026-101900: Defect that allows header injection in the fetch adapter via an inherited FormData getHeaders function.
CVE-2026-101901: Defect that allows an unhandled error event during HTTP/2 session initialization to crash the process, leading to denial of service.
CVE-2026-101902: Defect that could allow prototype pollution to override the HTTP method used by the default instance.
CVE-2026-101903: Defect that allows crafted data: URLs to trigger regular expression backtracking that freezes the event loop, leading to denial of service.
CVE-2026-101904: Defect that allows header injection via inherited headers after a minimal request interceptor.
CVE-2026-101905: Defect that could allow prototype pollution to hijack the request socket via an inherited createConnection function.
CVE-2026-101906: Defect that allows untrusted redirect locations to trigger quadratic host normalization during proxy bypass checks, leading to denial of service.
CVE-2026-101907: Defect that causes maxRedirects: 0 to be ignored by the fetch adapter, enabling redirect-based server-side request forgery.
CVE-2026-101908: Defect that could allow prototype pollution in the fetch adapter to alter outbound requests.
CVE-2026-101909: Defect that could allow prototype pollution to alter form data serialization options.
CVE-2026-62380: Defect that allows unvalidated address and credential fields in the SOCKS4 and SOCKS5 client encoders to enable null byte, CRLF, and credential injection.
CVE-2026-71290: Defect that causes the built-in TLS hostname verification policy to be ignored by the asynchronous HTTP client, enabling server impersonation by an attacker able to intercept traffic.
CVE-2026-64607: Defect that allows a content-encoding decode error to leak pooled connections, leading to connection pool exhaustion and denial of service.
CVE-2026-84939: Defect that allows a malformed locale identifier to enable path traversal during template loading.
CVE-2026-82617: Defect that allows untrusted text processed by the built-in email and URL name finders to trigger excessive backtracking and stack overflow, leading to denial of service.
CVE-2026-68497: Defect that allows unbounded number parsing when deserializing Duration and XMLGregorianCalendar values, leading to denial of service.
CVE-2026-83557: Defect that allows Comparable to be used as a base type for polymorphic deserialization, bypassing the intended type validation.
CVE-2026-19032: Defect that allows Path values to be deserialized without a scheme allowlist, enabling resolution through arbitrary file system providers.
CVE-2026-91776: Defect that retains every unknown type ID encountered during deserialization, leading to unbounded memory growth and denial of service.
CVE-2026-91777: Defect that allows crafted input with many forward object references to trigger quadratic processing time, leading to denial of service.
CVE-2026-53914: Defect that allows unsafe deserialization in the Kotlin build cache, enabling code execution.
CVE-2026-90852: Defect that allows a shared compression dictionary to be used after it is freed, leading to memory corruption.
CVE-2026-87824: Defect that allows oversized sample lengths during dictionary training to read past buffer boundaries, causing a JVM crash.
CVE-2026-90560: Defect that allows unvalidated offset and length arguments in the decompression dictionary constructor to read memory out of bounds, causing a JVM crash.
CVE-2026-87825: Defect that allows a dictionary to be closed while still referenced by a stream or context, leading to silent data corruption or a JVM crash.
CVE-2026-87795: Defect that allows unvalidated offset and length arguments in the compression dictionary constructor to read native heap memory out of bounds, causing a JVM crash.
CVE-2026-71889: Defect that causes X.509 name constraints not to be applied to the end-entity certificate during certificate path review.
CVE-2026-71888: Defect that allows the streaming CMS AuthenticatedData parser to accept messages with inconsistent digest algorithm and authenticated attribute fields, enabling MAC verification bypass.
CVE-2026-18036: Defect that allows secret-dependent division timing in NTRU operations to leak secret key material via a timing side-channel.
CVE-2026-18040: Defect that allows HQC encapsulation and decapsulation to leak secret-derived data through cache and timing side-channels.
CVE-2025-53864: Defect that allows deeply nested JSON in a JWT claim set to trigger uncontrolled recursion, leading to denial of service.
CVE-2026-93451: Defect that allows crafted compressed data to trigger a heap buffer overflow in the typed-array decompression methods.
CVE-2026-108104: Defect that allows framed data with a large declared chunk length to cause pooled buffers to be released twice, exposing or overwriting other streams' decompressed data.
CVE-2026-108106: Defect that allows an attacker-declared uncompressed length to trigger unbounded memory allocation, leading to denial of service.
CVE-2026-116590: Defect that allows unchecked ranges in array copy and raw write operations to crash the JVM.
CVE-2026-41715: Defect that allows the HTTP client to leak credentials when following a redirect that downgrades the protocol.
CVE-2026-47843: Defect that could allow a previously configured DNS resolver to be reused by clients with a different resolver configuration.
CVE-2026-47845: Defect that could cause the remote IP address to be evaluated incorrectly when HAProxy Protocol is enabled.
CVE-2026-47848: Defect that could allow the WebSocket client to leak credentials when a handshake is redirected to a different origin.
CVE-2026-47857: Defect that allows the windowTimeout operator with fair backpressure enabled to be exploited, leading to denial of service.
CVE-2026-47863: Defect that allows the bufferTimeout operator with fair backpressure enabled to be exploited, leading to denial of service.
CVE-2026-47874: Defect that allows HTTP/1.1 pipelined requests over a single connection to cause excessive memory consumption, leading to denial of service.
CVE-2026-59171: Defect in the org.json library used by the BigQuery connection.
CVE-2026-54399: Defect that allows crafted HTTP/1 headers to cause excessive memory consumption during parsing, leading to denial of service.
CVE-2026-54428: Defect that allows an unlimited HTTP/2 header list size before the settings are acknowledged, leading to denial of service.
CVE-2026-110985: Defect that causes the default SASL/OAUTHBEARER validator to fall back to a validator that does not verify JWT signatures when no JWKS endpoint is configured, enabling authentication bypass.
CVE-2026-89407: Defect that allows crafted numeric input to trigger quadratic regular expression backtracking, leading to denial of service.
CVE-2026-89425: Defect that allows invalid tokens to be reported without a length limit, leading to excessive memory consumption and denial of service.
Runtime:
CVE-2026-62380: Defect that allows unvalidated address and credential fields in the SOCKS4 and SOCKS5 client encoders to enable null byte, CRLF, and credential injection.
CVE-2026-71290: Defect that causes the built-in TLS hostname verification policy to be ignored by the asynchronous HTTP client, enabling server impersonation by an attacker able to intercept traffic.
CVE-2026-64607: Defect that allows a content-encoding decode error to leak pooled connections, leading to connection pool exhaustion and denial of service.
CVE-2026-84939: Defect that allows a malformed locale identifier to enable path traversal during template loading.
CVE-2026-82617: Defect that allows untrusted text processed by the built-in email and URL name finders to trigger excessive backtracking and stack overflow, leading to denial of service.
CVE-2026-68497: Defect that allows unbounded number parsing when deserializing Duration and XMLGregorianCalendar values, leading to denial of service.
CVE-2026-83557: Defect that allows Comparable to be used as a base type for polymorphic deserialization, bypassing the intended type validation.
CVE-2026-19032: Defect that allows Path values to be deserialized without a scheme allowlist, enabling resolution through arbitrary file system providers.
CVE-2026-91776: Defect that retains every unknown type ID encountered during deserialization, leading to unbounded memory growth and denial of service.
CVE-2026-91777: Defect that allows crafted input with many forward object references to trigger quadratic processing time, leading to denial of service.
CVE-2026-53914: Defect that allows unsafe deserialization in the Kotlin build cache, enabling code execution.
CVE-2026-90852: Defect that allows a shared compression dictionary to be used after it is freed, leading to memory corruption.
CVE-2026-87824: Defect that allows oversized sample lengths during dictionary training to read past buffer boundaries, causing a JVM crash.
CVE-2026-90560: Defect that allows unvalidated offset and length arguments in the decompression dictionary constructor to read memory out of bounds, causing a JVM crash.
CVE-2026-87825: Defect that allows a dictionary to be closed while still referenced by a stream or context, leading to silent data corruption or a JVM crash.
CVE-2026-87795: Defect that allows unvalidated offset and length arguments in the compression dictionary constructor to read native heap memory out of bounds, causing a JVM crash.
CVE-2026-71889: Defect that causes X.509 name constraints not to be applied to the end-entity certificate during certificate path review.
CVE-2026-71888: Defect that allows the streaming CMS AuthenticatedData parser to accept messages with inconsistent digest algorithm and authenticated attribute fields, enabling MAC verification bypass.
CVE-2026-18036: Defect that allows secret-dependent division timing in NTRU operations to leak secret key material via a timing side-channel.
CVE-2026-18040: Defect that allows HQC encapsulation and decapsulation to leak secret-derived data through cache and timing side-channels.
CVE-2025-53864: Defect that allows deeply nested JSON in a JWT claim set to trigger uncontrolled recursion, leading to denial of service.
CVE-2026-93451: Defect that allows crafted compressed data to trigger a heap buffer overflow in the typed-array decompression methods.
CVE-2026-108104: Defect that allows framed data with a large declared chunk length to cause pooled buffers to be released twice, exposing or overwriting other streams' decompressed data.
CVE-2026-108106: Defect that allows an attacker-declared uncompressed length to trigger unbounded memory allocation, leading to denial of service.
CVE-2026-116590: Defect that allows unchecked ranges in array copy and raw write operations to crash the JVM.
Azure Connections:
CVE-2026-41715: Defect that allows the HTTP client to leak credentials when following a redirect that downgrades the protocol.
CVE-2026-47843: Defect that could allow a previously configured DNS resolver to be reused by clients with a different resolver configuration.
CVE-2026-47845: Defect that could cause the remote IP address to be evaluated incorrectly when HAProxy Protocol is enabled.
CVE-2026-47848: Defect that could allow the WebSocket client to leak credentials when a handshake is redirected to a different origin.
CVE-2026-47857: Defect that allows the windowTimeout operator with fair backpressure enabled to be exploited, leading to denial of service.
CVE-2026-47863: Defect that allows the bufferTimeout operator with fair backpressure enabled to be exploited, leading to denial of service.
CVE-2026-47874: Defect that allows HTTP/1.1 pipelined requests over a single connection to cause excessive memory consumption, leading to denial of service.
Google BigQuery:
CVE-2026-59171: Defect in the org.json library used by the BigQuery connection.
Databricks:
CVE-2026-54399: Defect that allows crafted HTTP/1 headers to cause excessive memory consumption during parsing, leading to denial of service.
CVE-2026-54428: Defect that allows an unlimited HTTP/2 header list size before the settings are acknowledged, leading to denial of service.
Kafka Connection:
CVE-2026-110985: Defect that causes the default SASL/OAUTHBEARER validator to fall back to a validator that does not verify JWT signatures when no JWKS endpoint is configured, enabling authentication bypass.
Ignition Module:
CVE-2026-89407: Defect that allows crafted numeric input to trigger quadratic regular expression backtracking, leading to denial of service.
CVE-2026-89425: Defect that allows invalid tokens to be reported without a length limit, leading to excessive memory consumption and denial of service.
