@ -8,7 +8,7 @@ When it comes to using AI there are different approaches and different reasons f
- **No SDK**, most AI models allows you to interact directly with the AI model via for example HTTP requests. That approach works and may sometimes be your only option if an SDK option is missing.
- **SDK**. Using an SDK is usually the recommended approach as it allows you to type less code to interact with your model. It usually is limited to a specific model and if using different models, you might need to write new code to support those additional models.
- **A framework**. A framework usually takes things to another level in the sense that if you need to use different models, there's one API for all of them, what differs is usually the initial set up. Additionally frameworks brings in useful abstractions like in the AI space, they can deal with tools, memory, workflows, agents and more while writing less code. Because frameworks are usually opinionated they can really be helpful if you buy into how they do things but may fall short if you try to do something bespoke that the framework isn't made for. Sometimes a framework can also simplify too much and you may therefore not learn an important topic that later may harm perfomance for example.
- **A framework**. A framework usually takes things to another level in the sense that if you need to use different models, there's one API for all of them, what differs is usually the initial set up. Additionally frameworks brings in useful abstractions like in the AI space, they can deal with tools, memory, workflows, agents and more while writing less code. Because frameworks are usually opinionated they can really be helpful if you buy into how they do things but may fall short if you try to do something bespoke that the framework isn't made for. Sometimes a framework can also simplify too much and you may therefore not learn an important topic that later may harm performance for example.
Lee Stott
2 months ago
committed by
GitHub