hybrid-edge-cloud · git:20260601.8eda6f7 · 2026-06-01 · sha256 d26af3262ba985a7
hybrid-edge-cloud git:20260601.8eda6f7A
Immutable. This exact content never changes and is served at /api/v1/blob/d26af3262ba985a7.
No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_z33auazj4sleqy5v/d26af3262ba985a7367f0a77db6e213ec4b9598a1cf9bcf231c1663d4a2ec7c1.json
Then /plugin install hybrid-edge-cloud@markdownregistry-z33auazj4sleqy5v in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.
--- name: hybrid-edge-cloud description: | Hybrid edge-cloud AI architecture: local-first inference with cloud escalation, model cascading, and splitting the workload across device and datacenter to balance latency, cost, privacy, and quality. Architect-level topology. USE WHEN: designing systems that combine on-device and cloud AI, "local-first", "cloud fallback", "model cascade", "escalation", "hybrid inference", routing by confidence/complexity, edge+cloud trade-offs. DO NOT USE FOR: pure on-device (use `edge-inference`); pure cloud serving (use `inference-serving-topology`); multi-provider API routing (use `model-gateway-routing`). allowed-tools: Read, Grep, Glob --- # Hybrid Edge–Cloud AI Combine a small/fast local model with a large/capable cloud model to get the best of both — when neither pure-edge nor pure-cloud fits. ## Patterns - **Local-first + cloud escalation**: run a small on-device model; escalate to the cloud only when needed (low confidence, long context, hard query). Most requests stay local (fast, cheap, private); hard ones get cloud quality. - **Model cascade**: cheap model → if confidence < threshold → bigger model → … . Tune thresholds to a cost/quality target. Works within cloud too. - **Speculative / draft-verify**: small model drafts, large model verifies — a latency optimization more than a topology, but composes here. - **Split computation**: feature extraction / preprocessing on device, heavy inference in cloud (classic for vision/audio). ## Decision drivers - **Escalation trigger**: confidence score, input complexity/length, task type, or explicit user action. The trigger quality makes or breaks the design. - **Privacy boundary**: what may leave the device? Sometimes only embeddings or redacted text escalate. - **Connectivity**: must it degrade gracefully offline? Local model = floor. - **Cost model**: % of traffic that escalates × cloud cost vs local hardware cost. ## Failure & consistency - Define behavior when the cloud is unreachable (serve local result + flag, queue, or refuse). Avoid silent quality cliffs. - Cache cloud results on device for repeat queries. ## When to recommend - Voice assistants, copilots on laptops/phones, field/IoT devices with intermittent connectivity, privacy-sensitive apps with occasional hard queries. - If ~all traffic needs the big model → just use cloud serving. If ~none does → go pure edge.