Splunk Cloud Connect data management for the agentic SOC

This topic identifies the data flow and management for the various agents in Splunk Enterprise Security in the following categories:

  • Architecture and data inventory

  • Storage residency retention and deletion

  • Security access and customer controls

  • Network and connectivity architecture

  • AI models and action controls

  • Assurance and externally shareable evidence

  • Regional and regulatory approval requirements

  • Cloud Connect platform scope and architecture

Architecture and data inventory

Table 1. Sample data table
Data management category Detection builder agent Triage agent Guided Response agent Autonomous Response agent
Bi-directional data flow across the customer environment and Splunk Enterprise Security User prompts, user prompt file attachments, active chat session history, event-based detection metadata from only the detection currently being edited, DBA custom instructions on the stack, and metadata provided by the SPL Assistant.

Customer to cloud: Eligible findings are exported with their field value payload, (except Splunk internal fields whose names start with mv). Exporter adds the source system, schema version, dispatch time, and request identifiers. When configured, it also adds Mission Control notes and SOAR automation action-result messages. Investigation triage requests contain the investigation ID and qualifying finding IDs or timestamps.

Cloud to customer: Triage results are stored as a Mission Control AI analysis, including status, summary, justification, evidence or tools, hypotheses, suggested disposition and execution or trace identifiers. Triage does not provide a file-upload flow.

Customer to cloud: Prompts, conversation or page context, user or organization identifiers and authorization tokens; detection names, SPL and settings; findings, investigations, notes, response plans/tasks; user or asset attributes, risk or anomaly evidence and threat indicators; requested search text/time ranges and results; SOAR action parameters, status and results; and file content where uploads are enabled. Cloud to customer: answers, summaries, generated SPL, recommendations, approval requests and supported action/change requests. Authorized searches can return raw event text and any permitted field values.

Customer to cloud: the investigation ID and authoritative investigation record, response plan and tasks, serialized findings, notes, the existing Triage AI analysis, detection/configuration context, intermediate-finding overview, recent automation history/SOAR run details, analyst instructions or clarification replies, and results from authorized SPL searches or SOAR actions. Tenant tier/stack metadata and correlation IDs support routing. Cloud to customer: task status and bounded AI summaries, required task notes, approval/clarification prompts, investigation progress, search/action requests and results, and a Response-owned AI analysis/final execution record. The Response Agent does not enable file-upload tools. Caller credentials are removed before persistence and fresh service credentials are hydrated only for execution.
End-to-end architecture and data-flow Data is transferred from the client's browser to the customer's stack to Splunk Cloud for processing then to the customer's stack and finally to the client's browser for rendering. The ES/Mission Control modular input identifies findings from detections with AI triage enabled, enriches them when applicable, and writes them to a local dispatch queue. With activated Cloud Connect it initiates an outbound HTTPS request to the finding-export API using a Cloud Connect bearer token; the legacy path forwards a finding_triage stash through the configured ES backend route. The cloud triage service processes the work item and returns an analysis that Mission Control persists and presents. The stack also retains a local retry record until the export is accepted or reaches a terminal outcome. 1. The assistant sends the prompt and relevant page/conversation context to Splunk cloud processing. 2. The cloud requests authorized Splunk searches or Splunk/SOAR API operations; returned records go back to cloud processing for analysis. Cloud Connect carries requests/results for the connected Splunk environment. 3. Relevant prompts, conversation context and retrieved data are supplied to the selected model. The configured GPT-5.4 default is hosted by Microsoft Azure AI Foundry. 4. Answers, approvals and action requests are presented in the assistant. Approval audit events use the existing Cloud Connect connection. Conversation/workflow state and files persist in cloud storage; diagnostic/usage records can hold additional copies. Mission Control launches the asynchronous investigations.soc_analyst agent with an investigation ID and tokenless tenant context. Agent Runtime loads the current investigation, response plan/tasks, findings, notes, Triage analysis and relevant automation/detection context; the model plans and executes tasks in strict order. Bounded SPL and configured SOAR operations use fresh tenant-scoped service credentials. The workflow writes task status/summaries/notes and a Response AI analysis back to Mission Control and persists execution/checkpoint state so it can pause for clarification or approval and resume. Connected Cloud uses Cloud Connect and the ES-owned SOAR proxy; the model provider is reached through Splunk's model route.
Data category and classification, in addition to any included identifying or sensitive information DBA has two categories of data: information as part of a LLM request and unique stack metada. Both categories are the minimum required information to provide accurate and useful responses: LLM Requests: {prompts, file attachments, chat session history, responses, custom user instructions} Stack Personalization: {log metadata provided by the SPL Assistant, event-based detection metadata} Finding fields, investigation identifiers, notes and automation results are needed to explain and prioritize the security event and recommend a disposition or next step. Identifiers and dispatch metadata support routing, deduplication and audit correlation. These records are customer security data and may contain personal or confidential information, including usernames, email addresses, host/IP/DNS values, asset owners, free text and raw/matched event values. Generated triage summaries and evidence can repeat those values. Prompts and page/conversation context describe the task. Detection SPL/settings support authoring and explanation. Findings, search results, entity/risk data and threat indicators support investigation. Response plans, action parameters and results support response. Identity/authorization information enforces customer and user permissions. These transfers include customer security content and can contain personal or confidential data: names, email, employment/location details, usernames, host/IP/DNS/MAC values, asset owners, raw event text and investigation notes. Generated answers can repeat that information. Investigation, finding, detection and Triage data ground the response plan and prevent unsupported actions. Notes and analyst replies provide decision context. SPL results supply evidence; SOAR parameters/results support configured remediation. Identifiers and execution metadata support routing, recovery and audit. This is sensitive customer security data and can contain personal/confidential values such as names, email, usernames, host/IP/DNS/MAC data, asset owners, raw event text, notes and action targets. Generated summaries, evidence and action results can repeat those values.
Transfer triggers, run frequency, and turning off automatic data flow Transfers are triggered by user prompts. Only the relevant information required to provide an accurate and useful response are submitted on the user prompt. There is no automatic transfer of data. All AI features can be disabled in the Global AI settings. Transfer is automatic for new qualifying findings produced by a detection whose AI triage setting is enabled. The dispatcher runs every 30 seconds by default and sends bounded batches from the local queue; it may wait for configured post-triage automation context before enrichment. Customers can disable all AI features, disable triage for an individual detection, or disable the associated connector/action policy. There is no per-finding prompt before the initial triage export. Interactive transfers occur when a user submits a prompt, when the agent requests data, and when a task resumes after approval or another user response. Automated workflows retrieve data after their trigger and can continue after the initiating request ends. Customers can disable all AI features or specific autonomous agentic workflows. For Autonomous SOC Analyst agent, creation of findings from a detection is a trigger. A Response Agent run is started for an investigation/response-plan workflow, not by a periodic customer-data sweep. It proceeds through plan tasks, refreshes operational context at task boundaries, and resumes after an analyst clarification or required approval. Customers can disable the relevant AI/agentic capability, suppress SPL or SOAR execution with the Response passive-mode controls, disable individual connectors/actions, or require approval under customer action policy. A run already queued may continue until cancelled or reaches a terminal state, subject to those controls.
Limits to payload, event-samples, and transfer volumes. Only the relevant data is sent upon a user request. The maximum payload size is determined by the supported context window of the underlying model. Most payloads are below 16K tokens. Code defaults limit dispatch to 10 findings per detection run, 10 items per batch, and 200 findings in 24 hours; investigation triage is also capped at 200 investigations in 24 hours. The local queue defaults to a 1 GB maximum and queued items older than the configured 2-hour age are cleaned up, subject to retry handling. The finding sourcetype has TRUNCATE=0, so this is not a field/payload-size cap. No customer option for aggregate-only triage or for removing event fields by sampling is implemented in the reviewed exporter. There are limits placed on SPL search operation for assistant use-case. Search operation returns bounded result set, with a bounded time range. For uploaded files, configured limits are 50 MiB per file, 100 attachments and 200 MiB selected total. Search execution is bounded by earliest/latest time, result count and timeout; the preparer appends | head 100 when the SPL is otherwise unbounded and rejects an unqualified index=* scan. Runtime context uses bounded serializers and recent automation history (up to 50 entries in the documented adapter). The production model configuration caps generated output at 6,000 tokens. Exact total request-body/context-window limits are platform/model-route controls and should be confirmed. There is no Response setting that converts event results to aggregate-only data; analysts should use aggregate SPL when required.
Event samples obtained but raw security event data is not sent or retained DBA does not operate or process raw log samples. DBA is focused on detection authoring, tuning, and explaining. Triage does not export a separate archive of all raw security events. It does, however, export the finding/notable record almost in full: only __mv_* internal fields are removed. If the finding contains _raw, matched-event fields, original field values or analyst-entered notes, those values cross the boundary and may be included in the returned analysis. Therefore a blanket statement that no raw event data is sent is not accurate for Triage; the correct distinction is full event archives versus values embedded in the selected finding. Existing statements may be accurate for other workflows. Searches executed for agentic AI use-cases can transmit raw event text and field values, and returned data can persist in conversation records, working files and diagnostic traces. Response does not continuously export a raw event archive, but its authorized SPL searches can return raw event text and original field values. Those results are sent to cloud processing and can be represented in workflow checkpoints, task evidence/notes, action arguments and the final AI analysis. Therefore statements that no raw security-event data is sent or retained do not apply to Response; the correct boundary is the bounded result set of searches performed for the investigation.

Storage residency retention and deletion

Data management category Detection builder agent Triage agent Guided Response agent Response agent
Processing of received data (i.e. is data written to disk, cached, or held only in volatile memory) Processing – Requests are processed in memory by the Detection Builder Agent service in Splunk Cloud. The agent combines the user's request with the current detection configuration, applicable instructions and conversation history. It may also call tools to get more information. Proposed detection changes are applied in the editor after user approval and saved when the user clicks Save. Written to persistent storage – Conversation history, tool results and workflow state are written to a cloud database. Uploaded files, where supported, are checked and processed before the agent can use them. Source files and processed content are stored in cloud object storage. Tracing and telemetry may create additional records when enabled. See Data retention for how long each is kept. Cached – Attachment content may be cached in memory for the current turn to avoid repeated reads. Access is checked again on every operation. Conversation state is retained separately between requests. Credentials and tokens – Authentication credentials are handled separately from model-visible content. The service includes controls to exclude credentials from persisted execution requests and redact sensitive authentication fields from traces. LLM processing – Whether the model provider or gateway stores, caches or logs requests requires confirmation. On the customer stack, eligible records are copied, enriched and serialized to JSON. They can be written to the local dispatch/retry queue under $SPLUNK_HOME/var/run/splunk/ai_dispatch or to the legacy stash before outbound delivery. Cloud processing assembles the triage analysis and may use model context and workflow state; the returned result is written to Mission Control's mc_ai_analysis store. Credentials are separate from the finding payload. Exact cloud database, cache and model-provider storage behavior must be confirmed by the Triage service owner. The model processes the prompt, assembled conversation context and retrieved results to generate responses. The service writes prompts, replies, tool calls/results and workflow progress to persistent cloud storage for conversation continuity and recovery. Large results, saved history and uploaded/generated files can also be stored. Agent Runtime processes the assembled investigation context and tool results in memory, but the workflow is resumable rather than memory-only. It stores tokenless launch/execution records and LangGraph checkpoints in Postgres and writes task status/summaries/notes and the Response AI analysis to Mission Control. Operational context is refreshed at task boundaries. Caller tokens are discarded before persistence; fresh service credentials and SOAR tokens/endpoints remain process-local. Data-bearing SOC adapter trace spans suppress inputs/outputs, but workflow state still contains the data needed to resume. Response does not enable file-upload assets.
Data retention based on type, purpose, and length of time Conversation and agent state, including session instructions - Maintain conversation continuity and resume agent work, including after approvals - 14 days, removed by nightly cleanup. Uploaded files, including instruction files, where supported – Provide documents and instructions for the agent - 14-day configured default; production period requires confirmation. Disaster-recovery backups - Restore the service - Up to 7 additional days after deletion from active storage. LangSmith traces, when tracing is enabled - Diagnose execution; may include prompts, responses and tool activity - 30 days. User-submitted feedback telemetry - Improve quality - 90 days. Automatic Product Telemetry, when enabled - Measure product interactions and agent orchestration; may include chat messages and tool-call arguments - Retention period to be confirmed. Monitoring traces and metrics, excluding prompt and response content - Monitor service health - 8 days for traces; 6 months for metrics. Custom instructions stored in your Splunk deployment - Provide customer-specific guidance - Until a user changes or removes them. Copies sent to DBA as session instructions follow the applicable cloud retention periods above. Customer-stack queue files exist to support delivery and retry and are normally cleaned after processing or when they exceed the default two-hour queue age; pending direct-export retry state can remain while delivery is unresolved. The completed analysis is retained in Mission Control's AI analysis store under the applicable ES/Mission Control policy. The reviewed app-ess code does not establish retention periods for cloud execution state, model-provider processing, telemetry, backups or the Mission Control record. Those durations must be supplied by the Triage service, Platform, Privacy and Security owners before publication. Conversation records, including prompts, replies and tool results, support continuation/recovery. Working files support attachments, saved history and large results. Automatic database backups support recovery. Other retained data include completed workflow records, file metadata, diagnostic/usage records, etc. Any data can be retained up to 90 days. Default LangGraph checkpoint cleanup is 14 days after a workflow becomes terminal; queued/running executions are exempt until terminal. The agent_execution_results row stores tokenless launch context, status, result, internal context and timestamps, and its expiry is nullable; the reviewed implementation does not establish an automatic TTL for those rows. Mission Control task notes/status and Response AI analysis follow ES/Mission Control retention. Data-bearing SOC trace spans are redacted, and bounded operations metrics contain correlation/status/counter data. Exact production retention for execution rows, MC records, telemetry and backups must be supplied by Platform/Privacy before publication.
Configuration of retention periods and default values or minimum time period Customer configuration - Customers cannot configure retention periods for DBA data stored in Splunk Cloud. Splunk manages these periods. Defaults and minimums - Conversation and agent state have a default retention period of 14 days. Uploaded files, where supported, also have a configured 14-day default, subject to production confirmation. These are not guaranteed minimums. Customers cannot select a minimum or maximum. Customer-controlled instructions - Custom instructions remain in your Splunk deployment until a user changes or removes them. Copies sent to DBA as session instructions follow the cloud retention periods described under Data retention. Retention exceptions - Whether customers can request a different retention period requires confirmation from the service owner. The reviewed settings let customers control whether Triage runs, but do not expose customer-configurable retention periods for queued exports, cloud execution state or Mission Control AI analyses. Queue age is an operator/configuration default rather than a customer data-retention control. Any supported contractual exception or service-side retention override requires confirmation from the service owner. Customers cannot configure cloud retention periods. Retention is service-operated. Customers cannot configure Response checkpoint or execution-record retention in the reviewed product settings. threadRetentionDays is a service deployment setting, not a customer control, and Mission Control retention follows the platform/service policy. Customers can disable the workflow or action paths, but disablement is not equivalent to deleting existing records. Any supported retention exception requires the Platform/Privacy service owner.
Data deletion time and method and proof of data deletion Conversation and agent state - Records older than the retention period are deleted by a nightly database cleanup. Database maintenance then reclaims the storage space. Uploaded files - Removing a file, or reaching its expiry date, stops the agent from accessing it. The file content is then deleted in the background, either by a cleanup process or by storage lifecycle rules. Physical deletion may finish after access is removed. File metadata and audit records may remain afterwards. Custom instructions - Users can change or remove instructions in their Splunk deployment. Copies already saved in cloud conversation state or traces aren't deleted at that point. They're removed when the retention periods for those stores end. Backups, traces and telemetry - These copies expire automatically when their own retention periods end. Local queue/stash items are removed after successful or terminal processing and by age-based cleanup; retry state is preserved while an export remains pending. Disabling Triage stops new exports but does not itself delete completed Mission Control analyses or cloud-side copies. The reviewed feature code does not provide a customer deletion API, deletion certificate or proof-of-deletion event for Triage data. Cloud workflow state, telemetry, backups and Mission Control records follow their own service retention/deletion controls and require owner confirmation. Scheduled cleanup jobs delete records whose creation date is more than retention maximum. Terminal LangGraph checkpoints are removed by scheduled cleanup after the configured retention period (14 days by default); active queued/running workflows are retained so they can resume. The reviewed agent_execution_results records have no automatic expiry guarantee, and Mission Control task/AI-analysis records have separate lifecycle rules. Cancelling or disabling Response stops work but does not prove immediate deletion of all copies. No customer-facing deletion-certificate mechanism is defined in the reviewed code. Platform must specify deletion of execution rows, MC records, traces, telemetry and backups.
Persistence of data samples in logs, debug traces, caches, crash dumps, or backups Logs – Service logs record operational metadata. Error messages may occasionally include fragments of request content. Telemetry – When enabled, Product Telemetry can include chat messages and tool-call arguments, which may contain detection configuration, SPL or other customer-provided content. Monitoring traces and metrics don't include prompt or response content. Debug traces – When diagnostic tracing is turned on, traces may contain prompts, responses and tool activity, including customer data given to the agent. Backups – Disaster-recovery backups can keep copies of stored data for up to 7 days after it's deleted from active storage. Copies after deletion – Each copy follows the retention period of the system that holds it. Deleting the original doesn't immediately delete copies in telemetry, traces or backups. Yes. Before delivery, the complete finding plus any attached notes/automation messages can exist in local JSON queue, retry or stash files. Operational logs can contain request/finding identifiers, status and error details. The returned analysis persists in Mission Control. Whether raw payloads appear in cloud logs, traces, caches, crash dumps or backups is not defined by the customer-stack code and must be answered by the Triage service/Platform owners; deleting the local queue item does not prove deletion of those other copies. No, service logs, metrics and diagnostic data doesn't contain any customer data. Yes. Investigation context and tool results can persist in Agent Runtime execution rows/checkpoints and in Mission Control task notes/status and AI analysis records. The SOC data-bearing adapter spans are configured with payload_redacted=true, and operations-state telemetry is limited to correlation/status/counters rather than raw prompts or outputs. Credentials are excluded/redacted. The reviewed code does not establish backup/crash-dump retention or prove that no customer text can appear in sanitized errors; those residual-copy controls require Platform/Security confirmation.
Countries and regions through which the data pipeline runs The app-ess implementation does not select or document countries for Triage compute, model inference, execution storage, telemetry, backups or support access. It uses the customer's configured Cloud Connect route. Use the shared regional Cloud Connect/model-availability disclosure and have the Triage service owner confirm every processing and storage location, including cross-region failover and model-provider processing. Services are processing and persisting data in paired cloud-connected regions. If the default model is selected to be used then any model inference calls are made to Model Provider cloud (For ES AI use-cases it is Azure). Model/LLM processes prompts and supplied conversation/search context within these boundaries: US deployments use the US zone; Ireland, Germany, Italy and France use the European zone; India, Indonesia, South Korea, Singapore, Australia and Japan use the APAC zone. Brazil, the UK and Canada use global processing. Processing can leave the selected hosting region and, within Europe/APAC, cross country borders. The Response code selects a model configuration but does not define the production countries for Agent Runtime compute, Mission Control storage, model inference, telemetry, backups or support access. Use the shared Cloud Connect and regional model-availability disclosure. The existing platform answer identifies Microsoft Azure AI Foundry for the default GPT route; Platform/Privacy must confirm the actual route and every processing/storage location for the customer's region, including cross-region failover.

Security access and customer controls

Control category Detection builder agent Triage agent Guided Response agent Response agent
Masking, redaction, anonymization, or pseudonymization before data leaves your environment DBA only processes user submitted data and event-based detection metadata. It does not process sensitive data requiring masking, redaction, anonymization, or pseudoanonymization unless the user submits it. The only automatic field removal in the reviewed export path is deletion of fields whose names begin with __mv_. There is no general masking, anonymization or pseudonymization of finding values, _raw content, notes or SOAR automation messages before they leave the customer environment. Customers should control sensitive content through the data generated by the detection and through enabled triage/automation settings; the reviewed code does not provide a per-field exclusion list. There is no common automatic masking or anonymization step before agent data leaves the customer environment. Search results and investigation records can include original event values, names, email addresses, IP addresses and free text. Credentials, authorization containers and recognized secret keys are removed or redacted before workflow persistence/tracing, and fresh credentials are hydrated only for tool execution. There is no general masking or anonymization of the investigation, finding, notes, SPL results or SOAR action content before it leaves the customer environment; those values are required for analysis. Customers can reduce exposure by limiting the investigation content, search scope, result fields and enabled actions.
Permissions that govern cloud-requested data access and searches on premises The SPL Assistant may execute queries to validate the syntax and semantics of recommended SPL. The stack administator may configure whether the SPL Assistant can run queries on the stack. Triage is an automated system workflow, not an interactive search running as the initiating analyst. The local modular input uses its Splunk service/session context to read eligible findings, notes and automation history; the outbound request is authenticated with the tenant's Cloud Connect token. Access is therefore governed by the configured ES service identity, detection eligibility and Mission Control/SOAR integration permissions rather than the permissions of a particular end user. The minimum role/capability list should be confirmed in the deployment documentation. Interactive searches through Cloud Connect use the requesting user's Splunk identity and its permissions. Automated SOC investigation and SOAR response workflows use separate splunk identity created for AI agent execution. The roles and permissions of the identity are pre-configured and are visible from Splunk User/Role management. Autonomous Response operations do not impersonate the analyst who launched the run. Agent Runtime uses a fresh server-owned, tenant-scoped ES credential for authorized Splunk access. In Connected Cloud, SOAR calls use the ES-owned SOAR proxy rather than persisting or minting a direct SOAR credential; direct-cloud mode obtains a tenant SOAR endpoint/token through the approved redirection flow. Customer connector/action/asset configuration and policy govern SOAR execution. Approval audit delivery uses an analyst request-scoped token only when one is live; the runtime does not substitute another identity.
Audits on data transfers, stored objects, requested searches, and cloud-side access Requested searches – Searches the agent runs use your session and permissions in your Splunk environment. What you can audit depends on your environment's normal search auditing, permissions and retention settings. User approvals – Where approval auditing is enabled on a supported ES version, approval and rejection decisions are sent to your Splunk _audit index. Each record includes the tool name, correlation identifiers, a timestamp and redacted decision details. Delivery is best effort, so some decisions may be missing. Stored objects – Custom instructions and saved detections are stored in your Splunk deployment under your Enterprise Security permissions and change tracking. These local records don't list the copies held in cloud systems. Session identifiers – Each conversation and run has identifiers that Splunk Support can use to look into a specific session. Data transfers and cloud-side access – Your local search and approval records show specific actions. They aren't a complete audit of data sent to the cloud or of access to copies stored there. Records of cloud-side processing and access are kept in Splunk's internal systems. Customers can audit local dispatcher logs and normal Splunk/Mission Control records for finding/request identifiers and delivery outcomes. Mission Control AI analysis records include status, timestamps, owner, evidence/tools, execution ID, trace/run identifiers and any recorded response actions; underlying finding and automation history remain auditable under their normal ES/SOAR controls. The reviewed code does not expose a separate complete ledger of every cloud-side read or model access, and best-effort operational logs are not a deletion attestation. Agent approval audit events which include the operation, decision, timestamps and redacted inputs/responses are published to customer's audit index on Splunk. These events are delivered through the established Cloud Connect connection. Customers can review response-plan task status and bounded summaries, required task notes, investigation progress, the Response AI analysis (summary, justification, evidence, tools, actions and execution/trace identifiers), and the underlying Splunk search/SOAR automation history. Agent Runtime also records lifecycle, checkpoint/interrupt and execution status for recovery/operations. HITL approval audit delivery is available when a live analyst token is present; durable async runs without one do not fabricate an audit identity, so no approval audit event is published in that case. Cloud-side model/access auditing remains a Platform control.
Data encryption in transit and at rest, and management of encryption keys Tenant isolation: DBA runs on shared Splunk Cloud infrastructure and is designed to keep each customer's data logically separate. Customers don't get a dedicated database, server or model. Searches and tools: Searches use your session and permissions. Results go only to the conversation that requested them. Conversation state: Conversation and agent state is designed to be kept separate for each customer and conversation. Influence on outputs: Responses are built from your request, your detection, your conversation and data from your environment. Splunk-provided instructions are the same for all customers and are shipped with the app. Direct Triage export uses HTTPS with bearer authentication and disallows redirects. The legacy route uses the configured Splunk forwarding/Cloud Connect transport. At-rest encryption, key ownership/rotation and encryption of cloud databases, traces and backups are platform controls rather than properties established by the feature code. Use the shared Platform/Security encryption statement for those layers. Stored conversation data and files are encrypted at rest. File encryption uses service-managed keys. Database connections require encrypted transport, and configured model and diagnostic endpoints use HTTPS. Response traffic uses the shared HTTPS/TLS Cloud Connect/service paths; direct service calls use authenticated TLS. Credentials are kept separate from persisted model-visible content. At-rest encryption, KMS/key ownership, rotation and encryption of Postgres, Mission Control, telemetry and backups are not established by the Response feature code; use the shared Platform/Security encryption and key-management answer for those layers.
People and services that can access your data, required approvals, and visibility of access DBA complies with Splunk's General Terms, Splunk's Specific Terms, Cisco's Privacy Statement, and Cisco's Responsible AI Terms for customer data access, approvals, and visibility. Service access includes the ES/Mission Control dispatcher, Cloud Connect or the legacy backend-forwarding service, the Triage cloud service/model route, and Mission Control storage for the result. When enabled, Mission Control and SOAR services supply notes and automation history. The reviewed code does not identify human support or engineering access, break-glass procedures or model-provider personnel access; those require the shared Platform/Privacy access-control response. Personnel access to production customer data is restricted to support scenarios and governed by privileged-access management controls. Service access includes Agent Runtime, the configured model route, Mission Control/ES context APIs and storage, the shared SPL/SAIA components used to generate or run bounded searches, and SOAR/ES proxy services for configured actions. Fresh tenant-scoped service credentials authorize tool access. The reviewed feature code does not identify human support/engineering access or model-provider personnel access; Platform/Privacy must provide roles, approval, logging and break-glass controls.
Tenants isolation, i.e. can one user's data impact another user's outputs Tenant isolation: DBA runs on shared Splunk Cloud infrastructure and is designed to keep each customer's data logically separate. This doesn't mean each customer gets a dedicated database, server or model. Searches and tools: Searches use your session and permissions and leave out detections you can't access. Results go to the conversation that requested them. Conversation state: Conversation and agent state is designed to stay separate for each customer and conversation. Influence on outputs: Responses are built from your request, the current detection, the instructions that apply, your conversation history and the tool results the agent is authorised to get. DBA is designed to keep other customers' private data out of that context. Each export uses tenant-scoped Cloud Connect configuration/token material and carries identifiers needed to route the work item back to the correct ES/Mission Control tenant. The customer-side code does not prove the cloud service's database, cache, queue, trace or model-gateway isolation controls. Platform and Triage service owners must provide the authoritative logical-isolation and cross-tenant access testing statement. Data processing at service layer is stateless and there are tenant isolation and entitlement access checks that validate the customer organization and conversation owner. Execution records and credential hydration are tenant-scoped, and runtime requests carry the tenant/stack context required to bind reads and writes to the correct Mission Control and SOAR environment. Caller tokens are not used as a cross-tenant key. The feature code alone does not prove all database row-level, cache, checkpoint, trace or model-gateway isolation controls; Platform/Security must provide the authoritative isolation design and validation evidence.
Issue, storage, rotation, and revocation of tenant service credentials for authentication and authorization Authentication: Each request uses a token issued by your Splunk deployment for the signed-in user. Splunk Cloud services authenticate separately, using Splunk-managed service identities. Authorization: The agent acts with your Splunk permissions. The model can't choose or change the credentials used for a request. Cloud Connected deployments: Splunk issues a tenant-scoped service token for the relay connection, separate from your user token. If the required credentials aren't available, the request fails. It doesn't fall back to a more privileged identity. Storage: Credentials are kept separate from content the model sees. They're designed to be kept out of stored conversation state, logs and telemetry. Rotation and revocation: User tokens expire and are renewed under your Splunk authentication settings. Splunk renews and rotates its own service credentials. Revoking access may not immediately cancel tokens that have already been issued. Local ES REST calls use the modular input's session/service context. Cloud delivery uses a Cloud Connect bearer token in the HTTPS Authorization header; the token is not part of the finding JSON. Issuance, secure storage, rotation, revocation and compromise response for that tenant credential are shared Cloud Connect controls. Revoking/disablement should stop new exports, but it does not delete data already delivered. Interactive requests use the user's signed Splunk token that are propagated for agent execution in context of the user and the tokens are not persisted anywhere. Ephemeral token for customer-scoped identity created for autonomous agents is periodically transmitted and synced with service managing agents. Caller authentication material is stripped before the async launch request is persisted. Before initial or resumed tool execution, Agent Runtime hydrates a fresh server-owned tenant Splunk credential. Direct-cloud SOAR credentials are obtained through the ES soar_redirection flow; Connected Cloud uses the ES-owned SOAR proxy and does not persist a direct SOAR token. Tokens/endpoints remain process-local and are excluded from checkpoints/logs/traces. Issuance, rotation, revocation and leaked-credential response follow the shared tenant credential lifecycle.

Network and connectivity architecture

Control category Detection builder agent Triage agent Guided Response agent Response agent
Required domains, URLs, ports, protocols, and regional endpoints Cloud Connected deployments: Allow outbound HTTPS (TCP 443) from search heads to api.scs.splunk.com, auth.scs.splunk.com, and your deployment's regional and tenant scs.splunk.com hostnames. Users' browsers also need to reach your tenant's scs.splunk.com hostname. Proxies: Supported if they allow streaming responses. Inbound access: Not required. Model providers: You don't need to allow them. DBA reaches models through Splunk Cloud services. Regions: DBA uses the endpoints for the region you selected during Cloud Connect setup. Before publishing, confirm: the Relay WebSocket endpoint, the chat UI bundle domain, exact hostnames vs *.scs.splunk.com, the minimum TLS version and whether TLS inspection works, which regions serve DBA and how failover works, and whether Splunk Cloud stacks need any allowlisting. Triage uses the shared Cloud Connect HTTPS path on TCP 443 and posts to /seca-v1-api/v1/seca-v1/finding-exports on the configured service host. Customers do not connect directly to the model provider. The legacy path forwards the finding_triage sourcetype through the esbagroup/backend route. Publish the shared api.scs.splunk.com, auth.scs.splunk.com and tenant/regional hostname allowlist from the Cloud Connect documentation; confirm the exact regional service hostname before release. Allow outbound traffic to *api.scs.splunk.com and *auth.scs.splunk.com on TCP port 443 for the documented shared Cloud Connect connection. https://help.splunk.com/en/splunk-enterprise-security-8/user-guide/8.8/introduction/access-splunk-cloud-connect-in-splunk-enterprise-security-to-access-cloud-extensions Response uses the shared Cloud Connect/service endpoints over outbound HTTPS/TCP 443; customers do not allowlist the model provider directly. Connected Cloud SOAR operations traverse the existing ES/Cloud Connect proxy path rather than a new customer-facing endpoint. Publish the shared api.scs.splunk.com, auth.scs.splunk.com and tenant/regional hostnames from the supported ES version, plus any documented SOAR prerequisite. The Response code does not define an additional inbound port.
Outbound and inbound connections Bidirectional activity is required. The customer environment initiates the Cloud Connect/export connection outbound. Requests and returned results are logically bidirectional over the established service path, but Triage does not require a separate unsolicited inbound network connection to the customer stack. Any legacy forwarding exception should be described with the shared Cloud Connect architecture. For Cloud Connect, the customer environment initiates the connection outbound. Requests, results and approval audit events travel in both directions over that established connection. Approval auditing does not require a separate inbound connection to the customer environment. The customer environment establishes Cloud Connect outbound. Investigation context, search/action requests, results, approvals and updates are bidirectional messages over established authenticated service paths; Response does not require a separate unsolicited inbound network connection to the customer stack. Direct Cloud versus Connected Cloud changes the SOAR routing path, not this network-establishment principle.
Supported proxy configuration and validation HTTP and HTTPS proxies are supported. The instructions are available in the Splunk Documentation (https://help.splunk.com/en/splunk-enterprise-security-8/user-guide/8.7/introduction/access-splunk-cloud-connect-in-splunk-enterprise-security-to-access-cloud-extensions). The direct exporter applies the Splunk proxy configuration to its HTTPS request. Customers should configure and validate the proxy through the supported Cloud Connect procedure, including DNS, TCP 443, authentication, streaming/long-lived connection support where required, certificate inspection compatibility and a successful finding-export health test. Triage does not define an independent proxy setting. See Supported proxy configurations. Response relies on the supported shared Cloud Connect proxy configuration and the ES-owned SOAR proxy in Connected Cloud; it has no independent end-user proxy setting. Customers should validate DNS and TCP 443 reachability, proxy authentication, TLS/certificate inspection compatibility, Cloud Connect health, a bounded SPL operation and—if enabled—a non-destructive SOAR test through the documented setup flow. Confirm version-specific prerequisites in the Cloud Connect/Response setup documentation.
Configuration of certificate and authentication settings so that Cloud-connected features use their existing user identities and permissions. Verification of the setup of new and upgraded connections Token-signing certificate: On customer-managed Splunk Enterprise, set certFile (an RSA PEM file containing both the certificate and its private key) and sslPassword under [oauth2_settings] in authentication.conf. Then reload token keys or restart Splunk. The default server.pem works, and the certificate doesn't need to come from a public certificate authority. User identity and permissions: DBA acts as the signed-in user, with that user's Splunk and ES permissions. If the user token isn't available, the request fails. It doesn't fall back to a more privileged identity. New connections: Cloud Connect shows as enabled, and the DBA chat is available in the detection editor. Upgrades: Check the certificate setting and DBA availability after upgrading. If the earlier assistant still appears, DBA setup isn't complete yet. Permission check: We recommend testing with a restricted user to confirm DBA only sees what that user can access. Triage has no separate end-user token-signing certificate or user-impersonation setup: it is a system-scoped workflow authenticated through the configured Cloud Connect connection. Customers should complete the shared Cloud Connect certificate/authentication setup, confirm the connection is enabled after installation or upgrade, enable AI triage for a test detection, and verify a successful export and returned AI analysis. Transport/proxy certificates are distinct from user-identity signing certificates. See Certificate authentication and settings. Complete the shared Cloud Connect token-signing certificate and authentication setup for the applicable ES deployment. Response itself uses a system-scoped tenant credential for autonomous operations rather than the interactive user's identity, so successful UI sign-in alone is not sufficient. Verify the Cloud Connect connection after installation/upgrade, the Response feature and tenant metadata, passive-mode settings, service identity/roles, SOAR proxy/redirection prerequisites, paired SOAR version metadata and a test workflow that pauses/resumes and writes its AI analysis. Identity-signing and transport/proxy certificates are separate controls.

AI models and action controls

Control category Detection builder agent Triage agent Guided Response agent Response agent
Models used Splunk offers recommended and self-hosted models. Recommended models are frontier models within Splunk's Cloud boundary. Self-hosted models are hosted directly by Splunk within its Cloud boundary. The current models are available in the Splunk Documentation (https://help.splunk.com/en/splunk-enterprise-security-8/administer/8.8/agentic-capabilities-in-splunk-enterprise-security/agentic-ai-offerings-in-splunk-enterprise-security). The reviewed Triage client code does not name a model or model operator. It supports the global use_internal_models setting, so model selection must honor the customer's internal-model-only policy. The Triage service owner must provide the current production model name/version, hosting/operator and regional route; pair that answer with the published ES agentic model availability documentation. ES AI use-cases default to GPT-5.x OpenAI models hosted by Microsoft Azure AI Foundry. The internal-only default is gpt-oss-120b. The current production Response configuration uses GPT-5.4 with low reasoning effort and a 6,000-token output cap. It honors the internal-model-only request/policy when that option is selected. The code identifies the model but not the commercial hosting/operator or processing region; the existing shared ES answer identifies Microsoft Azure AI Foundry for the default GPT route. Platform should confirm the exact production operator, version/alias and regional route at publication time.
Opting out of providing training data for fine-tuning, evaluation, research and development, analytics, and cross-customer improvement. Splunk requires explicit customer consent (ie opt-in) for training, fine-tuning, or cross-customer improvement. More details are available in Splunk's General Terms, Splunk's Specific Terms, Cisco's Privacy Statement, and Cisco's Responsible AI Terms for customer data access, approvals, and visibility. The Triage feature code uses customer data to produce the requested triage analysis; it does not implement model training or cross-customer learning. That code fact is not a contractual guarantee about model-provider use, evaluation, product analytics or retained telemetry. Apply the shared Global AI setting and Splunk/Cisco Responsible AI and Specific Terms response, clearly separating model training consent from operational telemetry and service improvement, and have Privacy/Legal confirm the opt-out behavior. For the ES AI Assistant, published Global AI settings make training/fine-tuning use of AI service data off by default. Administrators can change that setting. Details on data usage can be found on Splunk terms - https://www.splunk.com/en_us/legal/splunk-specific-terms.html The Response implementation uses customer data to execute the requested investigation workflow and does not itself implement training or cross-customer learning. Its operational metrics are bounded status/counter/latency data, and data-bearing SOC adapter trace payloads are suppressed; that is distinct from model-provider data-use terms, evaluation datasets or product analytics. Apply the shared Global AI setting and Splunk/Cisco Responsible AI and Specific Terms answer, and have Privacy/Legal confirm training/fine-tuning defaults, evaluation/R&D uses and opt-out behavior for the configured model route.
Time duration for data remaining active in AI context and context reset Context resets at the termination of a session. Triage is work-item oriented rather than a user chat session. Each finding or investigation request supplies the context needed for that analysis; the returned AI analysis is a durable Mission Control record. The app-ess code does not define the cloud agent's active context/checkpoint lifetime or prove that context is erased at a model context-window boundary. Those cloud-state durations and reset semantics require the Triage service owner. Earlier messages and retrieved results are saved so a conversation or interrupted workflow can resume. Ending a session or reaching the context-window limit and subsequently summarizing context does not delete earlier messages. Response context is resumable across task boundaries, clarification and human-approval pauses. The initial Triage record is checkpointed; operational investigation context is refreshed at task boundaries. Reaching a model context-window boundary does not delete the workflow record. Default checkpoints remain for 14 days after terminal completion, while queued/running executions are exempt; execution rows and Mission Control outputs have separate lifecycles. Ending a UI session therefore does not reset or erase the Response workflow context.
Independent configuration of data-use and security controls None Customers can independently control the global AI enablement, internal-model-only selection, AI triage enablement per detection, preferred response template, and enabled connector/action policies. Post-triage settings include per-connector/action enablement, preferred/default asset, a customer human-approval prompt, exception matching, and allow-lists for promote/close dispositions. Telemetry opt-in is a separate setting. Defaults and version availability should be verified for the customer's ES release. Customers can disable all AI features or specific autonomous agentic workflows. ES Global AI settings separately control training/fine-tuning use of AI service data and model choice. Customers can separately control global AI availability and internal-model-only use; enable/disable the Response agentic workflow; set SPL and SOAR passive modes; enable specific SOAR connectors/actions/assets; require HITL under the customer action policy; configure exceptions; and constrain promote/close behavior with disposition allow-lists. The runtime fails closed when required configuration is unavailable. Settings, defaults and availability vary by ES/stack version and should be documented for the customer's release.
Run searches or actions autonomously or with manual approval When a user requests requires SPL, DBA will call the SPL Assistant which may execute the SPL on a limited time range to validate the syntax and semantics for the user's stack. DBA requires customer approval prior to authorizing the SPL Assistant to validate the query. Initial triage analysis is automatic for eligible findings and does not require per-finding approval. The dispatcher itself does not execute arbitrary SPL. Triage output can recommend a disposition and can participate in configured post-triage automation/action workflows. Customers can disable those workflows, limit enabled connectors/actions/assets and require the configured human-approval prompt; promote/close behavior is additionally constrained by disposition allow-lists. Yes. Agents can execute searches and, for enabled response workflows, submit security actions. Customers have ability to configure actions that agent execute with human approval. More details here - https://help.splunk.com/en/splunk-enterprise-security-8/administer/8.7/ai-assistant-in-security-and-agentic-capabilities/setting-up-the-ai-soc-analyst-agentic-workflow-in-splunk-enterprise-security Yes. Response can execute bounded read-only SPL searches and configured SOAR actions while processing response-plan tasks, including after the initiating request ends. Operations are sequential and grounded to the investigation/plan. Manual approval follows customer policy, but hard-coded risky actions, policy exceptions and unknown matches require approval/break-glass; the agent also pauses for clarification when required inputs are missing. Passive modes can suppress SPL or SOAR execution without removing all AI analysis.
Restrictions on validating searches or actions requested from the cloud DBA does not perform autonomous or independent searches without user request or consent. Controls include per-detection triage enablement, required finding identifiers, per-run and 24-hour limits, queue limits, tenant-authenticated HTTPS delivery, and redirect refusal. Downstream response actions are restricted by enabled connector/action configuration, preferred/default asset, exception rules, approval policy and disposition allow-lists. The exporter does not accept arbitrary cloud-supplied SPL for local execution. For interactive use-case any SPL execution is bounded by user's access and user's ad-hoc search capacity limits. SPL is restricted by a shared read-only parser, tenant service permissions, explicit earliest/latest/count/timeout bounds, rejection of unqualified index=* scans and an added | head 100 when otherwise unbounded. SOAR uses configured connectors/actions/assets; exact attached actions are preferred, parameters are schema/argument validated, and policy/HITL determines execution. Fresh tenant credentials are hydrated only at execution, and failures do not fall back to a more privileged identity. Response plan/task grounding and bounded evidence IDs restrict what the model can request.
Guardrails against prompt injection in attacker-controlled data Enterprise Security AI features include AI guardrails to detect and investigate Finding fields, notes and automation messages are untrusted and may contain attacker-controlled text. The customer-side exporter does not perform prompt-injection detection or sanitize that text; it removes only __mv_* fields. Security therefore depends on the cloud Triage agent treating the payload as data, limiting tools/actions, validating outputs and applying action/HITL policy. Those cloud guardrails and known limitations are not established in app-ess and must be documented by the Triage service owner before publication. There are several deterministic and model driven guardrails in place - validated tool arguments, limited available operations, source-system permissions, a read-only command allowlist for the shared SPL tool, bounded result retrieval, approval for selected searches/actions etc. Findings, notes, event fields and action results are treated as untrusted evidence, not authority. Deterministic controls constrain the workflow to investigation candidates and ordered response-plan tasks, validate evidence/citation mappings and tool arguments, restrict SPL to a read-only allowlist with bounded results, and restrict SOAR to configured actions/assets. Customer policy and mandatory approval for risky/exception cases gate side effects. Credentials are unavailable to the model and are hydrated only by the executor. These controls reduce but do not eliminate prompt-injection risk; model-generated text is not itself execution authorization.

Assurance and externally shareable evidence

Control category Detection builder agent Triage agent Guided Response agent Response agent
Third-party processors or sub-processors, their role, and location Subprocessors: See the applicable Offer Disclosure on the Cisco Trust Portal for approved subprocessors, what they do and where they process data. Model processing: DBA reaches AI models through Splunk-managed cloud services to process requests and run agent functions. Hosting: Cloud infrastructure providers run the service's compute, storage and backups. Diagnostic tracing: Execution traces are stored on Splunk-hosted infrastructure. Processing locations: Hosting, model processing, backups and support access may happen in different locations. The Offer Disclosure lists them. No Triage-specific third party is identified in the customer-stack code. Triage relies on the shared Splunk cloud/Cloud Connect hosting chain and the platform-configured model route; Mission Control and SOAR also participate when their context is included. Use the applicable Cisco Trust Portal Offer Disclosure for approved cloud, model, telemetry and support subprocessors, roles and locations. Privacy/Legal and the Triage service owner must validate the final list. AWS hosts the reviewed conversation database and file storage. Microsoft Azure AI Foundry hosts GPT-5.x and receives the prompts and conversation/retrieved context supplied for inference. Its processing boundary is described in Q12. No additional Response-specific subprocessor is identified in the reviewed code. Response uses the shared Splunk Agent Runtime/Mission Control hosting, the configured model provider, and ES/SAIA/SOAR services; the existing shared answer identifies AWS-hosted service storage and Microsoft Azure AI Foundry for the default GPT route. Use the applicable Cisco Trust Portal Offer Disclosure for the authoritative provider roles and processing locations, and have Privacy/Legal validate the list for the customer's region and model selection.
Service-specific data-use, privacy, and contractual terms
  • https://www.splunk.com/en_us/about-splunk/splunk-data-security-and-privacy/responsible-ai.html
  • https://www.cisco.com/c/en/us/about/legal/privacy-full.html https://www.splunk.com/en_us/legal/splunk-general-terms.html
  • https://help.splunk.com/en/splunk-enterprise-security-8/release-notes-and-resources/8.6/splunk-enterprise-security-release-notes/release-notes-for-splunk-enterprise-security#c58fb1c7b-f26e-430d-b071-5248f65998ac--en
  • https://www.splunk.com/en_us/legal/splunk-specific-terms.html
  • https://trustportal.cisco.com/c/r/ctp/trust-portal.html?language=English&search_keyword=splunk
No separate contractual terms are implemented by the Triage client. Apply the Splunk General Terms, Splunk Specific Offering Terms, Cisco Privacy Statement, Cisco Responsible AI terms/materials and the applicable Offer Disclosure/Trust Portal documentation. Privacy and Legal should confirm that the selected terms expressly cover Triage, its model route, retained analyses and any enabled SOAR automation. Splunk General Terms - https://www.splunk.com/en_us/legal/splunk-general-terms.html Specific Offering Terms - https://www.splunk.com/en_us/legal/splunk-specific-terms.html No separate contract is embedded in the Response implementation. Apply the Splunk General Terms, Splunk Specific Offering Terms, Cisco Privacy Statement, Cisco Responsible AI materials and the applicable Trust Portal Offer Disclosure. Privacy/Legal should confirm that the selected terms cover Agent Runtime execution/checkpoint storage, Mission Control records, the configured model route, SPL processing and any enabled SOAR actions.
Process for determining a tenant or service compromise, customer notifications, and notification timeline
  • Detection: DBA runs on Splunk Cloud services, which are monitored for signs of compromise and unauthorised access. DBA logs authentication, authorisation and service errors to support investigations. Response: Suspected incidents are handled through Splunk's security incident response process.
  • Notification: For incidents that require customer notification, Splunk contacts your designated Security Contact. Timeline: Notification obligations and deadlines follow your agreement with Splunk.
Triage agent uses the shared Splunk cloud security monitoring and incident-response process; the feature code does not define a separate notification channel or deadline. Splunk would investigate affected service/tenant activity and notify the customer's designated security contact when required. The governing contract and applicable law determine the notification timeline. Security Incident Response and Legal must confirm customer-specific requirements such as CERT-In reporting. Response agent uses the shared Splunk cloud monitoring and security incident-response process; the agent code does not define a feature-specific customer notification channel or deadline. Splunk would investigate affected tenant/service activity and notify the customer's designated security contact when required. Contractual commitments and applicable law determine timing. Security Incident Response and Legal must confirm customer-specific requirements, including any six-hour or regulator-specific deadline.
Security assessments, certifications, and regulatory assessment boundaries for the selected services, regions, and versions

Regional and regulatory approval requirements

Category Detection builder agent Triage agent Guided Response agent Response agent
Audit rights and regulatory support available to regulated financial customers in each applicable country or region
  • Audit rights: Your agreement sets your audit rights, including any Financial Services Addendum or Security Audit Addendum. Don't assume that general platform audit rights cover DBA. Audits don't include direct access to production systems.
  • Regulatory support: The Financial Services Addendum covers regulatory cooperation, subcontracting and exit assistance. Commitments depend on your contracted offering and jurisdiction. Your account team can confirm the terms that apply to you.
  • Evidence: Platform compliance reports, certificates, security questionnaires and pen-test affirmation letters can support your due diligence. Some require an NDA. They aren't DBA-specific evidence.
  • DBA scope: Confirm that your agreement and audit scope explicitly include DBA and the cloud and model-processing services it relies on. Connecting DBA to a regulated Splunk deployment doesn't extend that deployment's audit coverage to DBA.
Triage agent does not add feature-specific audit or regulator-access rights in code. Available audit rights, regulatory cooperation, evidence, subcontractor controls and exit assistance are governed by the customer's agreement and any Financial Services/Security Audit Addendum. Legal and Compliance should confirm that the contractual and assurance scope expressly includes Triage, Cloud Connect, the model route and enabled SOAR dependencies for the customer's jurisdictions. Autonomous response agent does not create feature-specific audit/regulator-access rights in code. Rights, regulatory cooperation, evidence, subcontractor controls and exit assistance are governed by the customer's agreement and any Financial Services/Security Audit Addendum. Legal and Compliance should confirm that the relevant assurance and audit scope includes Agent Runtime, Mission Control, the model route and enabled SOAR services for the customer's jurisdiction; connecting Response to an assessed deployment does not automatically extend that assessment boundary.
Splunk's contractual commitment to storage, processing, backups, and support access within a customer region Country and region availability. Country and region availability Country and region availability Country and region availability
AWS regions supported through Splunk Cloud Connect
  • [US West (Oregon)] us-west-2;
  • [US East (N. Virginia)] us-east-1;
  • [Europe (London)] eu-west-2;
  • [Europe (Ireland)] eu-west-1;
  • [Europe (Frankfurt)] eu-central-1;
  • [Europe (Milan)] eu-south-1;
  • [Europe (Paris)] eu-west-3;]
  • [Asia Pacific (Jakarta)] ap-southeast-3;
  • [Asia Pacific (Mumbai)] ap-south-1;
  • [Asia Pacific (Seoul)] ap-northeast-2;
  • [Asia Pacific (Singapore)] ap-southeast-1;
  • [Asia Pacific (Sydney)] ap-southeast-2;
  • [Asia Pacific (Tokyo)] ap-northeast-1;
  • [Canada (Central)] ca-central-1;
  • [South America (São Paulo)] sa-east-1
  • [US West (Oregon)] us-west-2 ;
  • [US East (N. Virginia)] us-east-1;
  • [Europe (London)] eu-west-2 ;
  • [Europe (Ireland)] eu-west-1 ;
  • [Europe (Frankfurt)] eu-central-1;
  • [Europe (Milan)] eu-south-1;
  • [Europe (Paris)] eu-west-3;
  • [Asia Pacific (Jakarta)] ap-southeast-3 ;
  • [Asia Pacific (Mumbai)] ap-south-1 ;
  • [Asia Pacific (Seoul)] ap-northeast-2 ;
  • [Asia Pacific (Singapore)] ap-southeast-1 ;
  • [Asia Pacific (Sydney)] ap-southeast-2 ;
  • [Asia Pacific (Tokyo)] ap-northeast-1 ;
  • [Canada (Central)] ca-central-1 ;
  • [South America (São Paulo)] sa-east-1
  • [US West (Oregon)] us-west-2 ;
  • [US East (N. Virginia)] us-east-1 ;
  • [Europe (London)] eu-west-2 ;
  • [Europe (Ireland)] eu-west-1 ;
  • [Europe (Frankfurt)] eu-central-1 ;
  • [Europe (Milan)] eu-south-1 ;
  • [Europe (Paris)] eu-west-3 ;
  • [Asia Pacific (Jakarta)] ap-southeast-3 ;
  • [Asia Pacific (Mumbai)] ap-south-1;
  • [Asia Pacific (Seoul)] ap-northeast-2 ;
  • [Asia Pacific (Singapore)] ap-southeast-1 ;
  • [Asia Pacific (Sydney)] ap-southeast-2 ;
  • [Asia Pacific (Tokyo)] ap-northeast-1 ;
  • [Canada (Central)] ca-central-1 ;
  • [South America (São Paulo)] sa-east-1
  • [US West (Oregon)] us-west-2 ;
  • [US East (N. Virginia)] us-east-1 ;
  • [Europe (London)] eu-west-2 ;
  • [Europe (Ireland)] eu-west-1 ;
  • [Europe (Frankfurt)] eu-central-1 ;
  • [Europe (Milan)] eu-south-1 ;
  • [Europe (Paris)] eu-west-3 ;
  • [Asia Pacific (Jakarta)] ap-southeast-3 ;
  • [Asia Pacific (Mumbai)] ap-south-1;
  • [Asia Pacific (Seoul)] ap-northeast-2 ;
  • [Asia Pacific (Singapore)] ap-southeast-1 ;
  • [Asia Pacific (Sydney)] ap-southeast-2 ;
  • [Asia Pacific (Tokyo)] ap-northeast-1 ;
  • [Canada (Central)] ca-central-1 ;
  • [South America (São Paulo)] sa-east-1
Canonical documentation for your extension's Cloud Connect integration.