|
1 | | -# typechat-go |
| 1 | +# typechat-go |
| 2 | + |
| 3 | +typechat-go is a library that makes it easy to build natural language interfaces using types. |
| 4 | + |
| 5 | +This project is a Go implementation of the original project [TypeChat](https://github.com/microsoft/TypeChat) by Microsoft. |
| 6 | + |
| 7 | +Visit https://microsoft.github.io/TypeChat for more information on what it enables you do. |
| 8 | + |
| 9 | +This implementation loosely follows what version 0.10 of TypeChat does with slightly different ergonomics more appropiate for Go. |
| 10 | + |
| 11 | +Some of the key differences are that this library has less opinions about how you communicate with the LLM, and so as long as you provide a valid client you can easily use this. |
| 12 | + |
| 13 | +## How to use |
| 14 | + |
| 15 | +### Prompt + Return Type |
| 16 | + |
| 17 | +This functionality allows you to pass in a natural language prompt and the expected result type you wish the LLM to use when replying. For example: |
| 18 | +```go |
| 19 | +type Classifier struct { |
| 20 | + Sentiment string |
| 21 | +} |
| 22 | + |
| 23 | +ctx := context.Background() |
| 24 | + |
| 25 | +// Provide a model client that implements the required interface |
| 26 | +// i.e. Do(ctx context.Context, prompt string) (response []byte, err error) |
| 27 | +// This model can call to OpenAPI, Azure or any LLM. You control the transport. |
| 28 | +model := ... |
| 29 | + |
| 30 | +prompt := typechat.NewPrompt[Classifier](model, "Today is a good day!") |
| 31 | +result, err := prompt.Execute(ctx) |
| 32 | +if err != nil { |
| 33 | + ... |
| 34 | +} |
| 35 | + |
| 36 | +fmt.Println(result.Sentiment) // provided by the LLM |
| 37 | +``` |
| 38 | + |
| 39 | +You'll notice that this implementation is using Generics, so the result you get from the LLM is fully typed and able to be uused by the rest of your application. |
| 40 | + |
| 41 | +### Prompt + Program |
| 42 | + |
| 43 | +This functionality allows you to pass in a natural language prompt along with an interface of behavior that your application supports. The library will have the LLM generate a sequence of steps it deems necessary to accomplish a given task. |
| 44 | + |
| 45 | +```go |
| 46 | +type API interface { |
| 47 | + CreateTweet(message string) |
| 48 | + CreateLinkedInMessage(message string) |
| 49 | +} |
| 50 | + |
| 51 | +ctx := context.Background() |
| 52 | + |
| 53 | +// Provide a model client that implements the required interface |
| 54 | +// i.e. Do(ctx context.Context, prompt string) (response []byte, err error) |
| 55 | +// This model can call to OpenAPI, Azure or any LLM. You control the transport. |
| 56 | +model := ... |
| 57 | + |
| 58 | +prompt := typechat.NewPrompt[API](model, "I really need to tweet and post on my LinkedIN that I've been promoted!") |
| 59 | +program, err := prompt.CreateProgram(ctx) |
| 60 | +if err != nil { |
| 61 | + ... |
| 62 | +} |
| 63 | + |
| 64 | +// Program will contain the necessary invocations your application has to do with the provided API to accomplish the task as idenfitied by the LLM. |
| 65 | +program.Steps[0].Name == "CreateTweet" |
| 66 | +program.Steps[0].Args == []any{"I have been promoted!"} |
| 67 | + |
| 68 | +program.Steps[1].Name == "CreateLinkedInMessage" |
| 69 | +program.Steps[1].Args == []any{"I have been promoted!"} |
| 70 | + |
| 71 | +// You can build a program executor on top of this structure. |
| 72 | +``` |
| 73 | + |
| 74 | +## Contributing |
| 75 | + |
| 76 | +This library is under development and still requires more work to solidify the provided APIs so use with caution. A release will be done at some point in the near future. |
| 77 | + |
| 78 | +### TODOs |
| 79 | + |
| 80 | +- More tests |
| 81 | +- Provide an adapter package to create out of the box with OpenAI and Azure clients |
| 82 | +- Other use cases such as conversations |
| 83 | +- Figure out best way to stay in sync with the original TypeChat project |
0 commit comments