On-device AI
Cihaz üzerinde yapay zekâ
D3 · Prompts, context, RAG, tools and agents
On-device AI executes AI inference on a local device, such as a phone or computer.
Review status: 2026-12-05
Technical explanation
A compatible model and runtime use the device’s available compute and memory.
Conceptual boundaries
The term identifies where inference runs. It does not imply that every part of the application is offline or that no data is transmitted.
Provider-neutral example
Illustrative: a note-taking app generates a draft summary with a locally installed model.
Limitations
A supported runtime or a successful phone demonstration does not establish compatibility with every device.
Related concepts
Atomic claims and evidence
1.1Google AI Edge: LiteRT-LM
- Source
- Google AI Edge: LiteRT-LM
- Source role
- Authoritative source
- Exact locator
- Run LLMs on-device with LiteRT-LM; Cross-platform; Hardware accelerated
- Supported claim
- Google documents LiteRT-LM for running language models on devices across mobile and desktop platforms.
- Last verification
- Review due
- Scope limitation
- This is one runtime’s scope, not a promise about all AI applications.
2.1Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
- Source
- Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
- Source role
- Authoritative source
- Exact locator
- Abstract: 3.8B phi-3-mini, phone deployment, and 7B phi-3-small
- Supported claim
- The Phi-3 report describes a model small enough for phone deployment.
- Last verification
- Review due
- Scope limitation
- This is a particular model demonstration, not evidence that all small models fit all phones.