Blend IT operations with agentic capabilities, and you’ll have AgenticOps.
Cisco unveiled this umbrella brand in June 2025, pairing it with a showcase named AI Canvas. The promise: a collaborative workspace orchestrating the various instances of Cisco AI Assistant, with an interface dynamically generated from telemetry data. It would feature multiple entry points, starting with Meraki and Splunk, and extending to the forthcoming Cisco Cloud Control—a unified engine for network policies, security, and observability.*
The general availability of AI Canvas is still pending. Consequently, for now the principal realization of the AgenticOps promise is Cisco AI Assistant. Its diffusion across the Cisco ecosystem continues, with Catalyst Center now on the horizon (it is available in beta, on request).
OT, Branches, Datacenter… AgenticOps Is Quietly Gaining Ground in the Cisco Ecosystem
At the same time, agentic troubleshooting and optimization capabilities are landing on OT networks. On the campus/branch side, configuration recommendations are entering the scene, complementing improvements to the AI Packet Analyzer module.
Still on the AgenticOps front, within the datacenter domain, event correlation on Nexus One is slated for June. Earlier (in May), features were expected to begin appearing in Security Cloud Control, for the firewall.
Another novelty: an Experience Metrics component (in alpha) intended to bridge UX data to bolster troubleshooting. It relies on integrations with Apple, Intel, Samsung, and Zebra.
In-House SLMs and Related Tools
To drive its AgenticOps approach, Cisco developed a specialized model (the Deep Network Model). By mid-2025, it was said to be trained on 40 million tokens. The latest update puts the figure at 100 million.
Several other models are associated with this one. Cisco expanded its capabilities in this area by acquiring in November 2025 a company that originated an LLMOps platform. The American group also built components designed to accelerate machine-data processing: LAPSE and ACE.
LAPSE (Lightweight Autonomous Program Synthesis and Execution) structures telemetry: it converts it on the fly into an optimized schema for the task.
ACE (Analytics Context Engineering) promotes context engineering. It brings analytics into the reasoning loop of LLMs by using a subset of SQL and Bash tools anchored to a virtual filesystem mapped to the observability APIs.
This system is meant to enable any model to be fed with “hybrid views” tailored to its capabilities, while still allowing it to probe raw data. Initially conceived to enhance the Deep Network Model, today it operates as a standalone service.
Views, oriented along rows or columns (determined by an algorithm), default to JSON. Cisco also offers a custom format inspired by the TOON project.
* Meraki and ThousandEyes were the first data sources for AI Canvas. The service handles telemetry from third-party equipment if injected manually or via Splunk. Cisco envisions enabling partners to add capabilities via API.
