edge-inference
Edge / on-device AI inference architecture: running models on MCUs, NPUs, mobile, and mini-PCs; quantization for edge, TOPS/memory/energy budgets, TinyML, and the latency case for on-device vs cloud. Architect-level. USE WHEN: designing on-device/edge AI, "edge inference", "on-device", "NPU", "TinyML", "quantization", "Jetson", "Coral", "Hailo", local LLM on small hardware, offline/low-latency inference, energy-constrained ML. DO NOT USE FOR: cloud GPU serving (use `inference-serving-topology`); choosing datacenter accelerators (use `ai-hardware-selection`); RAG app code (use rag skills).
What this skill does
# Edge / On-Device Inference ## Why on-device (the decision drivers) - **Latency**: on-device token/inference in ~tens of ms vs 200–500ms cloud round-trip — decisive for voice, AR, control loops. - **Privacy / offline / cost**: data never leaves the device; works without connectivity; no per-call cloud cost. - **Cost of doing it**: tight memory/energy/thermal budgets; smaller models; more engineering. ## The hardware tiers (match model to silicon) | Tier | Silicon | Typical model | |---|---|---| | MCU / TinyML | Cortex-M + tiny NPU (sub-$1 class) | KB-MB models: keyword spotting, anomaly detection (TFLite Micro) | | Mobile / AI-PC | Phone NPU, laptop NPU (tens of TOPS) | Quantized 3–8B LLMs, vision | | Edge box | Jetson Orin/Thor, Coral, Hailo (40+ TOPS) | 7–13B LLMs, multi-camera vision | ## Architectural levers - **Quantization** is the key enabler: FP16 → INT8 → INT4 trades accuracy for memory/throughput/energy. Most edge LLMs run INT4/INT8. Validate accuracy loss. - **Model choice**: small instruct/distilled models (e.g. 3–8B class) and vision models sized to the NPU's memory bandwidth, not just TOPS. - **Runtime**: TFLite/LiteRT, ONNX Runtime, ExecuTorch, llama.cpp/Ollama, vendor SDKs (TensorRT for Jetson). Pick what targets the accelerator. - **Budgets**: state the TOPS, RAM, and energy-per-inference budget up front — they bound everything. Batch=1, KV-cache memory dominates LLM RAM. ## When to recommend edge vs cloud vs hybrid - Hard latency / offline / privacy / per-unit cost → on-device. - Large model / variable load / centralized updates → cloud serving. - Both needs → **hybrid** (small local model + cloud escalation): see `hybrid-edge-cloud`.
Related in Cloud & DevOps
appbuilder-action-scaffolder
IncludedCreate, implement, deploy, and debug Adobe Runtime actions with consistent layout, validation, and error handling. Use this skill whenever the user needs to add actions to an App Builder project, understand action structure (params, response format, web/raw actions), configure actions in the manifest, use App Builder SDKs (State, Files, Events, database), deploy and invoke actions via CLI, debug action issues, or implement patterns such as webhook receivers, custom event providers, journaling consumers, large payload redirects, action sequence pipelines, and Asset Compute workers. Also trigger when users mention serverless functions in Adobe context, action logging, IMS authentication for actions, or cron-style scheduled actions.
orchestrating-datacloud
IncludedSalesforce Data Cloud product orchestrator for connect→prepare→harmonize→segment→act workflows. Use this skill when the user needs a multi-step Data Cloud pipeline, cross-phase troubleshooting, or data space and data kit management. TRIGGER when: user needs a multi-step Data Cloud pipeline, asks to set up or troubleshoot Data Cloud across phases, manages data spaces or data kits, or wants a cross-phase sf data360 workflow. DO NOT TRIGGER when: work is isolated to a single phase (use the matching phase-specific skill), the task is STDM/session tracing/parquet telemetry (use observing-agentforce), standard CRM SOQL (use querying-soql), or Apex implementation (use generating-apex).
github-project-automation
IncludedAutomate GitHub repository setup with CI/CD workflows, issue templates, Dependabot, and CodeQL security scanning. Includes 12 production-tested workflows and prevents 18 errors: YAML syntax, action pinning, and configuration. Use when: setting up GitHub Actions CI/CD, creating issue/PR templates, enabling Dependabot or CodeQL scanning, deploying to Cloudflare Workers, implementing matrix testing, or troubleshooting YAML indentation, action version pinning, secrets syntax, runner versions, or CodeQL configuration. Keywords: github actions, github workflow, ci/cd, issue templates, pull request templates, dependabot, codeql, security scanning, yaml syntax, github automation, repository setup, workflow templates, github actions matrix, secrets management, branch protection, codeowners, github projects, continuous integration, continuous deployment, workflow syntax error, action version pinning, runner version, github context, yaml indentation error
sf-datacloud
IncludedSalesforce Data Cloud product orchestrator for connect→prepare→harmonize→segment→act workflows. TRIGGER when: user needs a multi-step Data Cloud pipeline, asks to set up or troubleshoot Data Cloud across phases, manages data spaces or data kits, or wants a cross-phase `sf data360` workflow. DO NOT TRIGGER when: work is isolated to a single phase (use the matching sf-datacloud-* skill), the task is STDM/session tracing/parquet telemetry (use sf-ai-agentforce-observability), standard CRM SOQL (use sf-soql), or Apex implementation (use sf-apex).
fabric-cli
IncludedUse this skill for Fabric.so CLI workflows with the `fabric` terminal command: diagnose/install/login, search or browse a Fabric library, save notes/links/files, create folders, ask the Fabric AI assistant, manage tasks/workspaces, generate shell completion, check subscription usage, produce JSON output, and use Fabric as persistent agent memory. Do not use for Microsoft Fabric/Azure/Power BI `fab`, Daniel Miessler's Fabric framework, Python Fabric SSH, Fabric.js, or textile/fashion fabric.
lark
IncludedLark/Feishu CLI skills: lark-cli operations for docs, markdown, sheets, base, calendar, im, mail, task, okr, drive, wiki, slides, whiteboard, apps, approval, attendance, contact, vc, minutes, event. Use when the user needs to operate Lark/Feishu resources via lark-cli, send messages, manage documents, spreadsheets, calendars, tasks, OKRs, deploy web pages, or any Feishu/Lark workspace operations.