This topic covers the software and model stacks that make AI practical on local and edge hardware in 2026: lightweight inference runtimes, model compilers and quantization flows, agent/ orchestration frameworks, and platform integrations that connect on‑device inference to decentralized infrastructure and data pipelines. Demand for local inference has risen because of latency-sensitive vision and control workloads, tighter data‑privacy regulation, and the availability of efficient open models and edge accelerators. Key categories include Edge AI Vision Platforms (real‑time sensor and video inference), Decentralized AI Infrastructure (federated/peer and localized orchestration), and AI Data Platforms (on‑device labeling, sync, and feedback loops). Representative tools illustrate how these pieces come together: LangChain for engineering stateful agentic applications and deployment workflows; Mistral AI providing efficient open foundation models plus enterprise production tooling oriented to privacy and governance; Tabby as an open, self‑hosted coding assistant with local model serving; Windsurf (formerly Codeium) and JetBrains AI Assistant as agentic, in‑IDE platforms that blend local and cloud inference; and Code Llama, StarCoder, and WizardLM families as code‑specialized or instruction‑tuned open models often used for on‑prem or edge inference. Amazon CodeWhisperer exemplifies counterpart commercial developer assistants evolving toward hybrid local/cloud operation. Practically, choosing a local/edge stack in 2026 means matching model format and optimization toolchains to target hardware, using multi‑model orchestrators for fallbacks and privacy boundaries, and integrating with data platforms to collect labeled feedback while keeping sensitive data local. The result is an ecosystem where open models, compact runtimes, and orchestration frameworks enable production AI across constrained, regulated, or disconnected environments.