A continuous-model platform: snapshots the whole network with pyATS, runs tiered AI agents over it (Ollama → Haiku → Sonnet → Opus on escalation), proposes fixes, executes only after a human approves them.
Per-device audits (like Gladius) tell you a single device is broken. They don't tell you the network is broken. Cross-device reasoning, historical change tracking, and 'should we shut this interface or re-route the traffic?' need a different shape: a digital twin that watches the whole estate continuously.
Pulled inventory from Grafana — single source of truth, no manual lists. Kopis generates the pyATS testbed file from whatever's monitored.
Snapshots flow through a LangGraph state machine: each node is an agent, the state is shared, and conditional edges route based on what the previous agent found.
Tiered models by job: Ollama for normalisation (free, mechanical), Haiku for triage and classification (fast, cheap, the workhorse), Sonnet for remediation reasoning, Opus only when topology agent confidence is below 70% — the 'phone a friend' tier.
Recommendations land in an approval queue. Web UI or Slack — same record. After execution, Kopis automatically takes a fresh snapshot to verify the change worked.
Gladius tells you a device has vulnerabilities. Kopis tells you the network has problems and which fix to ship first.
An autonomous Cisco security auditor. Tell it an IP. It SSHes in, runs hardening checks, cross-references NIST 800-53 an…
A multi-tab SSH terminal with an AI sidekick that watches your session and suggests the right command before you Alt-Tab…
A streaming chat UI for Cisco's DevNet Content Search MCP. Ask anything about Meraki or Catalyst Center APIs and get a t…