feat(tools): additional tools in SQL toolkit #212
Conversation
Signed-off-by: Enrico Toniato <2827496+tonxxd@users.noreply.github.com>
There was a problem hiding this comment.
Adding the file support (public providers) that internally gets converted to SQL adds extra complexity (initialize method, mutation properties, checks) unrelated to the core SQL functionality.
I would add a static (async) factory method to convert and initiate the SQLTool with the SQLite connection. Something like this.
static async fromFile(provider: PublicProvider, ...) {
// conversion
return new SQLTool(...)
}
| // const sqlTool = new SQLTool({ | ||
| // provider: "sqlite", | ||
| // examples: [ | ||
| // { | ||
| // question: "Get wild albums", | ||
| // query: "SELECT * FROM Album where Title = 'Restless and Wild' LIMIT 1", | ||
| // }, | ||
| // ], | ||
| // connection: { | ||
| // dialect: "sqlite", | ||
| // logging: false, | ||
| // storage: await fetch( | ||
| // "https://github.com/lerocha/chinook-database/releases/download/v1.4.5/chinook_sqlite.sqlite", | ||
| // ).then(async (response) => { | ||
| // if (!response.ok) { | ||
| // throw new Error("Failed to download Chinook database!"); | ||
| // } | ||
|
|
||
| // const dbPath = path.join(os.tmpdir(), "bee_chinook.sqlite"); | ||
| // const data = Buffer.from(await response.arrayBuffer()); | ||
| // await fs.promises.writeFile(dbPath, data); | ||
| // return dbPath; | ||
| // }), | ||
| // }, | ||
| // }); |
There was a problem hiding this comment.
Either remove or create another example.
| ...old, | ||
| defaults: { | ||
| instructions: | ||
| "You are the Bee 🐝 Data Agent! If the user asks about data questions, use the FlowPilot tool, passing the user natural language question as 'question' to the tool. Do not render ascii tables, the user will already see it externally.", |
There was a problem hiding this comment.
I would rather not use emojis.
| ]; | ||
| } | ||
|
|
||
| function deduplicateAndSortResults( |
There was a problem hiding this comment.
Consider leveraging functionality from remeda library.
| import { Sequelize } from "sequelize"; | ||
| import { beforeEach, expect, vi } from "vitest"; |
| } as const; | ||
|
|
||
| export class SQLTool extends Tool<JSONToolOutput<any>, ToolOptions, ToolRunOptions> { | ||
| public options: ToolOptions; |
There was a problem hiding this comment.
Can be removed as it is defined in the parent class.
| }); | ||
| } | ||
|
|
||
| public constructor(options: ToolOptions) { | ||
| super(options); | ||
| if (!options.connection.dialect) { | ||
| this.options = options; |
There was a problem hiding this comment.
Can be removed as it is defined in the parent class.
Which issue(s) does this pull-request address?
#182
Description
This feature proposal introduces a toolkit for SQL and data analysis, by enhancing current SQL tool and introducing few other concepts
It adds excel and csv support for SQLTool in order to perform advanced querying on these type of files.
It also adds the ability to provide few shot examples for the connect db, useful for bigger complex dbs.
Checklist
yarn lintoryarn lint:fixyarn formatoryarn format:fixyarn test:unityarn test:e2e