Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/toolkit-ai/toolkit-ai
AI-agents that automatically generate and use Langchain Tools and ChatGPT plugins
https://github.com/toolkit-ai/toolkit-ai
Last synced: 3 months ago
JSON representation
AI-agents that automatically generate and use Langchain Tools and ChatGPT plugins
- Host: GitHub
- URL: https://github.com/toolkit-ai/toolkit-ai
- Owner: toolkit-ai
- Created: 2023-04-06T18:37:37.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2023-04-19T01:07:23.000Z (over 1 year ago)
- Last Synced: 2024-07-01T00:33:53.421Z (4 months ago)
- Language: TypeScript
- Size: 272 KB
- Stars: 517
- Watchers: 17
- Forks: 33
- Open Issues: 4
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- awesome-chatgpt - toolkit-ai/toolkit-ai - AI-agents that automatically generate and use Langchain Tools and ChatGPT plugins (SDK, Libraries, Frameworks / JavaScript/Typescript)
README
# 🧰 Toolkit AI
Toolkit is an easy way to generate and use AI plugins. Generate code for 🦜 🔗 LangChain plugins by just describing what they should do.
You can also try out the hosted version at [toolkit.club](https://toolkit.club)
This project is brought to you by the team from [Pal](https://www.heypal.chat) - an easy way to make an AI assistant that knows about your product, and has access to a wide set of tools to help your customers & team get more done.
## Roadmap
- [x] Generate LangChain tools via the npm package
- [x] CLI-based usage
- [x] Agent auto-runs & continually improve the tool
- [ ] ChatGPT plugin definition & hosting
## Usage
You can import this package, and generate a tool from your code.
yarn
```js
yarn add @heypal/toolkit-ai
```
pnpm
```js
pnpm add @heypal/toolkit-ai
```
```typescript
import Toolkit from '@heypal/toolkit-ai';
// OpenAI API key can optionally be set here, or the underlying model
// will use the environment variable OPENAI_API_KEY
const toolkit = new Toolkit({
openAIApiKey: '',
});
(async () => {
const tool = await toolkit.generateTool({
name: 'Temperature Converter',
description:
'Converts a temperature from Farenheit, Celsius, or Kelvin to any other unit.',
});
console.log(tool.langChainCode);
})();
```
We've also made an easy library you can try out the tools you generate here:
[hey-pal/langchain-tools-demo](https://github.com/hey-pal/langchain-tools-demo)
### Sample Output of Generating a Tool
The output will be a Tool object, that will include an implementation of the LangchainCode. Below is an example output that was generated for a LangChain tool
Example LangChain Tool Code Output
```javascript
import { Tool } from 'langchain/agents';
import Ajv from 'ajv';
// The following is the actual code that will be
// run by the tool when it is called
function call({ value, fromUnit, toUnit }) {
let convertedValue;
if (fromUnit === "Fahrenheit") {
if (toUnit === "Celsius") {
convertedValue = ((value - 32) * 5) / 9;
} else if (toUnit === "Kelvin") {
convertedValue = ((value - 32) * 5) / 9 + 273.15;
} else {
convertedValue = value;
}
} else if (fromUnit === "Celsius") {
if (toUnit === "Fahrenheit") {
convertedValue = (value * 9) / 5 + 32;
} else if (toUnit === "Kelvin") {
convertedValue = value + 273.15;
} else {
convertedValue = value;
}
} else if (fromUnit === "Kelvin") {
if (toUnit === "Fahrenheit") {
convertedValue = ((value - 273.15) * 9) / 5 + 32;
} else if (toUnit === "Celsius") {
convertedValue = value - 273.15;
} else {
convertedValue = value;
}
}
return { convertedValue };
}
// This is a class that corresponds to the Langchain tool definition
// https://js.langchain.com/docs/modules/agents/tools/
// It validates the input & output against the schemas
// and then it calls the tool code
class TemperatureConverter extends Tool {
name = 'temperature-converter';
description = `Converts a temperature from Fahrenheit, Celsius, or Kelvin to any other unit. The action input should adhere to this JSON schema:
{{"type":"object","properties":{{"value":{{"type":"number","description":"The temperature value to be converted."}},"fromUnit":{{"type":"string","enum":["Fahrenheit","Celsius","Kelvin"],"description":"The unit of the input temperature value."}},"toUnit":{{"type":"string","enum":["Fahrenheit","Celsius","Kelvin"],"description":"The unit to which the temperature value should be converted."}}}},"required":["value","fromUnit","toUnit"]}}`;
ajv = new Ajv();
inputSchema = {
"type": "object",
"properties": {
"value": {
"type": "number",
"description": "The temperature value to be converted."
},
"fromUnit": {
"type": "string",
"enum": [
"Fahrenheit",
"Celsius",
"Kelvin"
],
"description": "The unit of the input temperature value."
},
"toUnit": {
"type": "string",
"enum": [
"Fahrenheit",
"Celsius",
"Kelvin"
],
"description": "The unit to which the temperature value should be converted."
}
},
"required": [
"value",
"fromUnit",
"toUnit"
]
};
outputSchema = {
"type": "object",
"properties": {
"convertedValue": {
"type": "number",
"description": "The converted temperature value in the desired unit."
}
},
"required": [
"convertedValue"
]
};
validate(data, schema) {
if (schema) {
const validateSchema = this.ajv.compile(schema);
if (!validateSchema(data)) {
throw new Error(this.ajv.errorsText(validateSchema.errors));
}
}
}
async _call(arg) {
let output;
try {
const input = JSON.parse(arg);
this.validate(input, this.inputSchema);
output = await call(input);
try {
this.validate(output, this.outputSchema);
} catch (err) {
throw new Error(`${err.message}: ${JSON.stringify(output)}`);
}
} catch (err) {
output = { error: err.message || err };
}
return JSON.stringify(output);
}
}
export default TemperatureConverter;
```
## Toolkit Iterate CLI
If you'd like to try a longer-running tool generation process that utilizes self-evaluating agents, you can use the `toolkit-iterate` CLI:
1. Ensure [Docker](https://docs.docker.com/engine/install/) is installed on your system
2. Install this package globally: `npm install -g @heypal/toolkit-ai`
3. Create a JSON file with your partial tool specification, something like:
```json
{ "name": "Adder", "description": "Adds two numbers together" }
```
4. Run the CLI:
```sh
toolkit-iterate -v --inputJson=/path/to/spec.json --outputJs=/path/to/output.js --openAIApiKey=xyz --serpApiKey=xyz
```
Your API keys can also be read from the environment, using the variables `OPENAI_API_KEY` and `SERP_API_KEY`.