Release Notes for Crosswork AI, Release 2.0

Ease of use

Configuration Drift Detection

Uses machine learning to learn typical configuration patterns and identify unintended changes or anomalies in network device configurations. This capability helps maintain network consistency and compliance by detecting deviations early and enabling intelligent review of configuration changes. You can run analysis on demand or schedule it to run later for continuous review. You can also provide feedback for anomalies that do not require further action, helping suppress similar anomalies in future results.

Supported with Crosswork Network Controller and NSO.

Ease of use

Toxic Factor Identification

Uncovers hard-to-spot patterns associated with network events by analyzing events and inventory data. The agent uses data-driven insights to proactively identify risk factors associated with link down and BGP session down events that can contribute to network failures, downtime, and instability. You can run analysis on demand or schedule it to run at defined intervals for continuous monitoring.

Supported with Crosswork Network Controller.

Ease of use

Deep Network Troubleshooting

Diagnoses complex network issues with agent-driven troubleshooting. The agent analyzes a natural language issue description, generates candidate root-cause hypotheses, and validates them using available network data. You can review, remove, or add custom hypotheses and continue the investigation through additional iterations when more analysis is needed.

Supported with Crosswork Network Controller and NSO.

Ease of use

Knowledge Graph

Provides a structured view of network entities and relationships, helping agentic use cases understand topology and dependencies. The Knowledge Graph provides a normalized representation of the managed network for the agent, allowing you to browse node types, attributes, relationships, schema, and entity instances. You can also run GraphQL queries to explore network context and troubleshoot agent outcomes. Data is populated via Data Retrieval Adapters (DRAs), which act as an abstraction layer to integrate with any controller using the Crosswork AI SDK. The Crosswork Network Controller and Crosswork Network Service Orchestrator DRAs are included by default.

Ease of use

Observability

Provides visibility into agent and service activity through logs, metrics, traces, and default dashboards. Use observability views to monitor agent execution, LLM requests, token usage, tool calls, users, and model-related metrics. This observability data can also be analyzed using external tools, such as Splunk Observability Cloud, by importing the OpenTelemetry stream from Crosswork AI.

Available for Crosswork AI agents and services.

API Experience

Developer APIs and SDKs

Provides REST APIs and Python SDKs for building, operating, and integrating AI-assisted network automation use cases. Developers can build and deploy custom agents, MCP servers, and DRAs, while programmatically accessing Crosswork AI services such as LLM management, retrieval context, the Knowledge Graph, observability, and platform configuration. The Agent Catalog provides a central location to onboard and manage these custom agents.

Cisco Crosswork Network Controller

7.2.1(SP)

Cisco Network Services Orchestrator

6.4.12

CSCww70674

When the maximum number of Deep Network Troubleshooting threads is reached, a new session may remain idle for several minutes and fail with an error. This occurs when the available resources for Deep Network Troubleshooting sessions are exhausted.

Delete completed or older Deep Network Troubleshooting sessions from AI Assistant or wait for active sessions to be completed. After the sessions are complete, obsolete pods are terminated automatically and resources become available for new sessions. You can monitor pod status from Administration > System > Pods.

CSCww74586

The Configuration Drift Detection job may fail when an LLM model is updated during an active Deep Network Troubleshooting investigation.

When the LLM model configuration is updated while a Deep Network Troubleshooting investigation is running, the existing pod is marked as draining and continues the active investigation using the previous LLM configuration. A new DNT pod is created for new requests using the updated LLM configuration. On a Large deployment profile, the additional DNT pod can consume resources required to start a Configuration Drift Detection pod.

CSCww74590

When Configuration Drift Detection runs inference on very large device configurations, the task may exceed the default FORESIGHT_AGENT_TIMEOUT value of 48 hours and fail. This issue has been observed with large configurations of approximately 48,000 lines.

Increase the FORESIGHT_AGENT_TIMEOUT value using the following command:
kubectl set env deployment/foresight FORESIGHT_AGENT_TIMEOUT=<value-in-seconds> -n cisco-crosswork-ai

CSCww74592

When Configuration Drift Detection training is run for an XRv9k logical group with a large number of devices, the task may remain in a running state even after training has completed.

CSCww81968

In rare cases, after adding or deleting an external observability exporter, the Observability deployment may stop running. When this occurs, the Observability UI is not accessible and Observability API requests may return HTTP 500 errors.

Delete and redeploy the Observability deployment with the default collector configuration:

kubectl delete deployment observability -n cisco-crosswork-ai

kubectl apply -f /data/firmwared/cisco-crosswork-ai-<YOUR_VERSION>/nap-image-specs/on-enable/foresight-observability.yaml

After the deployment restarts, it retrieves the latest stored configuration from the database and resynchronizes the collector configuration.

Release Notes for Crosswork AI, Release 2.0

The current document.

Cisco Crosswork AI Getting Started, Release 2.0

Describes how to deploy and set up Crosswork AI and integrate it with Crosswork Network Controller.

Cisco Crosswork AI Administration, Release 2.0

Describes how to manage Crosswork AI, including system settings, software, backup and restore, certificates, users and roles, authentication settings, and audit logs.

Cisco Crosswork AI Configuration Drift Detection, Release 2.0 - For CNC

Describes how to use the Configuration Drift Detection agent to identify and review configuration anomalies.

Cisco Crosswork AI Configuration Drift Detection, Release 2.0 - For NSO

Describes how to use the Configuration Drift Detection agent with Cisco Network Services Orchestrator.

Cisco Crosswork AI Toxic Factor Identification, Release 2.0

Describes how to use the Toxic Factor Identification agent to identify factors associated with network events and instability.

Cisco Crosswork AI Deep Network Troubleshooting, Release 2.0

Describes how to use Deep Network Troubleshooting to investigate network issues, evaluate hypotheses, and identify probable root causes.

Cisco Crosswork AI Knowledge Graph, Release 2.0

Describes how to explore network entities, attributes, relationships, and graph data using the Knowledge Graph Explorer.

Cisco Crosswork AI Observability, Release 2.0

Describes how to use observability data, including traces, logs, and metrics, to understand agentic flows and agent behavior.

Crosswork AI API documentation

Provides REST APIs and Python SDKs for building, operating, and integrating AI-assisted network automation use cases.