CorporaApp

Overview

This is a Google Apps Script library for managing the corpora of Gemini API.

Description

The semantic search opens up a new wind for finding the expected values. Recently, the APIs for managing corpora have been added to Gemini API. Ref When the corpora of Gemini API is used, the semantic search can be achieved. Ref However, when the corpora are tried to be used with Google Apps Script, the script is complicated cumbersome. To address this challenge, I have created a library for managing the corpora using Google Apps Script. Ref With this library, managing corpora becomes effortless, requiring only straightforward scripts.

IMPORTANT

On February 4, the version of Generative Language API is v1beta. Ref In this library, the endpoint of v1beta is used. When the v1 is released, I would like to update the endpoint in the library.

Library's project key

1XrAybct1KUwGcFrEZ9BOd5sa0SoHeQwGhOWkDOHki9lDFAX9OHlO03y_

Usage

1. Install library

In order to use this library, please install the library as follows.

Create a GAS project.
- You can use this library for the GAS project of both the standalone and container-bound script types.
Install this library.
- Library's project key is 1XrAybct1KUwGcFrEZ9BOd5sa0SoHeQwGhOWkDOHki9lDFAX9OHlO03y_.

2. Linking Google Cloud Platform Project to Google Apps Script Project for New IDE

In this case, you can see how to do this at my repository.

Also, please enable Generative Language API at the API console.

After the above setting, the following sample script can be used.

Scopes

This library uses the following 2 scopes.

https://www.googleapis.com/auth/script.external_request
https://www.googleapis.com/auth/generative-language.retriever

Methods

Methods	Description
setAccessToken	Use own access token. For example, you want to use the access token from the service account. Please set this.

createCorpus	Create a new corpus.
deleteCorpus	Delete a corpus.
getCorpora	Get corpora list.
searchQueryFromCorpus	Search chunks from a corpus.
createDocument	Create a new document in a corpus.
deleteDocument	Delete a new document from a corpus.
getDocuments	Get document list from a corpus.
searchQueryFromDocument	Search chunks from a document.
setChunks	Put chunks into a document.
deleteChunks	Delete chunks from a document.
getChunk	Get a single chunk.
getChunks	Get chunk list from a document.
updateChunks	Update chunks in a document.
createPermission	Create permission of corpus.
deletePermission	Delete permission of corpus.
getPermissions	Get permission list of corpus.
updatePermission	Update permission of corpus.
searchQueryWithGenerateAnswer	Search chunks using models.generateAnswer.

In this library, the auto-completion can be used for the returned objects.

setAccessToken

Use own access token. For example, you want to use the access token from the service account. Please set this.

Default access token is retrieved by ScriptApp.getOAuthToken().

const accessToken = "###"; // Your access token
const res = CorporaApp.setAccessToken(accessToken).getCorpora();
console.log(res);

For example, if you want to use the access token retrieved from the service account. Please include the following scopes.
- https://www.googleapis.com/auth/generative-language.retriever
- https://www.googleapis.com/auth/script.external_request

createCorpus

Create a new corpus. Ref

const res = CorporaApp.createCorpus({ displayName: "sample" });
console.log(res.getContentText());

When this script is run, the following result is obtained.

{
  "name": "corpora/sample-###",
  "displayName": "sample",
  "createTime": "2024-01-01T00:00:00.000000Z",
  "updateTime": "2024-01-01T00:00:00.000000Z"
}

corpora/sample-### of name is the resource name of the created corpus.

deleteCorpus

Delete a corpus. Ref

CorporaApp.deleteCorpus("corpora/sample-###", false);

"corpora/sample-###" is the resource name of corpus. In this case, no value is returned.

About the 2nd argument of false, when this is false, when the document has chunks, the document is not deleted by occurring an error. The default value is false.

getCorpora

Get corpora list. Ref

const res = CorporaApp.getCorpora();
console.log(res);

When this script is run, the following result is obtained.

[
  {
    "name": "corpora/sample-###",
    "displayName": "sample",
    "createTime": "2024-01-01T00:00:00.000000Z",
    "updateTime": "2024-01-01T00:00:00.000000Z"
  },
  ,
  ,
]

searchQueryFromCorpus

Search chunks from a corpus. Ref

const res = CorporaApp.searchQueryFromCorpus(
  "corpora/sample-###",
  { query: "sample", resultsCount: 5 }
);
console.log(res.getContentText());

"corpora/sample-###" is the resource name of corpus. When this script is run, the following result is obtained.

{
  "relevantChunks": [
    {
      "chunkRelevanceScore": 0.6174245,
      "chunk": {
        "name": "corpora/sample-###/documents/sample-document-###",
        "data": {
          "stringValue": "sample value"
        },
        "customMetadata": [
          {
            "key": "key1",
            "stringValue": "value1"
          }
        ],
        "createTime": "2024-01-01T00:00:00.000000Z",
        "updateTime": "2024-01-01T00:00:00.000000Z",
        "state": "STATE_ACTIVE"
      }
    },
    ,
    ,
  ]
}

createDocument

Create a new document in a corpus. Ref

const res = CorporaApp.createDocument(
  "corpora/sample-###",
  {displayName: "sample document"}
);
console.log(res.getContentText());

"corpora/sample-###" is the resource name of corpus. When this script is run, the following result is obtained.

{
  "name": "corpora/sample-###/documents/sample-document-###",
  "displayName": "sample document",
  "createTime": "2024-01-01T00:00:00.000000Z",
  "updateTime": "2024-01-01T00:00:00.000000Z"
}

deleteDocument

Delete a new document from a corpus. Ref

CorporaApp.deleteDocument("corpora/sample-###/documents/sample-document-###", false);

"corpora/sample-###/documents/sample-document-###" is the resource name of document. In this case, no value is returned.

About the 2nd argument of false, when this is false, when the document has chunks, the document is not deleted by occurring an error. The default value is false.

getDocuments

Get document list from a corpus. Ref

const res = CorporaApp.getDocuments("corpora/sample-###");
console.log(res);

"corpora/sample-###" is the resource name of corpus. When this script is run, the following result is obtained.

[
  {
    "name": "corpora/sample-###/documents/sample-document-###",
    "displayName": "sample document",
    "createTime": "2024-01-01T00:00:00.000000Z",
    "updateTime": "2024-01-01T00:00:00.000000Z"
  },
  ,
  ,
]

"corpora/sample-###/documents/sample-document-###" is the resource name of the document.

searchQueryFromDocument

Search chunks from a document. Ref

const res = CorporaApp.searchQueryFromDocument(
  "corpora/sample-###/documents/sample-document-###",
  { query: "sample", resultsCount: 5 }
);
console.log(res.getContentText());

"corpora/sample-###/documents/sample-document-###" is the resource name of document. When this script is run, the following result is obtained.

{
  "relevantChunks": [
    {
      "chunkRelevanceScore": 0.6174245,
      "chunk": {
        "name": "corpora/sample-###/documents/sample-document-###",
        "data": {
          "stringValue": "sample value"
        },
        "customMetadata": [
          {
            "key": "key1",
            "stringValue": "value1"
          }
        ],
        "createTime": "2024-01-01T00:00:00.000000Z",
        "updateTime": "2024-01-01T00:00:00.000000Z",
        "state": "STATE_ACTIVE"
      }
    },
    ,
    ,
  ]
}

setChunks

Put chunks into a document. Ref

const resourceNameOfdocument = "corpora/sample-###/documents/sample-document-###";
const res = CorporaApp.setChunks(
  resourceNameOfdocument,
  {
    requests: [{
      parent: resourceNameOfdocument,
      chunk: {
        data: { stringValue: "sample value" },
        customMetadata: [{ key: "key1", stringValue: "value1" }]
      }
    }]
  }
);
console.log(res.map(r => JSON.parse(r.getContentText())));

"corpora/sample-###/documents/sample-document-###" is the resource name of document. When this script is run, the following result is obtained.

[
  { "chunks": [{ "name": "corpora/sample-###/documents/sample-document-###/chunks/###", "data": { "stringValue": "sample value" }, "customMetadata": [{ "key": "key1", "stringValue": "value1" }], "state": "STATE_ACTIVE" }] },
  ,
  ,
]

corpora/sample-###/documents/sample-document-###/chunks/### is the resource name of chunk.

It seems that in the current stage, the maximum size of value in metadata is 256 bytes. When this is over, an error like string_value cannot be more than 256 characters long. occurs. Please be careful about this.

deleteChunks

Delete chunks from a document. Ref

const res = CorporaApp.deleteChunks(
  "corpora/sample-###/documents/sample-document-###",
  { requests: [{ name: "corpora/sample-###/documents/sample-document-###/chunks/###" }] }
);
console.log(res.getContentText());

corpora/sample-###/documents/sample-document-### is the resource name of document.
corpora/sample-###/documents/sample-document-###/chunks/### is the resource name of chunk.
In this case, no value is returned.

getChunk

Get a single chunk. Ref

const res = CorporaApp.getChunk("corpora/sample-###/documents/sample-document-###/chunks/###");
console.log(res);

This method returns HTTPResponse.

getChunks

Get chunk list from a document. Ref

const res = CorporaApp.getChunks("corpora/sample-###/documents/sample-document-###");
console.log(res);

corpora/sample-###/documents/sample-document-### is the resource name of document. When this script is run, the following result is obtained.

[
  { "name": "corpora/sample-###/documents/sample-document-###/chunks/###", "data": { "stringValue": "sample value" }, "customMetadata": [{ "key": "key1", "stringValue": "value1" }], "state": "STATE_ACTIVE" },
  ,
  ,
]

corpora/sample-###/documents/sample-document-###/chunks/### is the resource name of chunk.

updateChunks

Update chunks in a document. Ref

const res = CorporaApp.updateChunks(
  "corpora/sample-###/documents/sample-document-###",
  {
    requests: [{
      chunk: {
        name: 'corpora/sample-###/documents/sample-document-###/chunks/###',
        // data: { stringValue: 'sample value' },
        customMetadata: [{ key: "add_key", stringValue: "Add value" }]
      },
      updateMask: "customMetadata"
    }]
  }
);
console.log(res.map(r => r.getContentText()));

corpora/sample-###/documents/sample-document-### is the resource name of document.
corpora/sample-###/documents/sample-document-###/chunks/### is the resource name of chunk.

When this script is run, the following result is obtained. The custom metadata is updated.

{
  "chunks": [
    {
      "name": "corpora/sample-###/documents/sample-document-###/chunks/###",
      "data": {
        "stringValue": "sample value"
      },
      "customMetadata": [
        {
          "key": "add_key",
          "stringValue": "Add value"
        }
      ],
      "createTime": "2024-01-01T00:00:00.000000Z",
      "updateTime": "2024-01-01T00:00:00.000000Z",
      "state": "STATE_PENDING_PROCESSING"
    }
  ]
}

createPermission

Create permission of corpus. Ref

const res = CorporaApp.createPermission(
  "corpora/sample-###",
  {
    granteeType: "USER",
    emailAddress: "###email address###",
    role: "READER"
  }
);
console.log(res.getContentText());

"corpora/sample-###" is the resource name of corpus. When this script is run, the following result is obtained.

{
  "name": "corpora/sample-###/permissions/###",
  "granteeType": "USER",
  "emailAddress": "###email address###",
  "role": "READER"
}

corpora/sample-###/permissions/### is the resource name of created permission.

deletePermission

Delete permission of corpus. Ref

const res = CorporaApp.deletePermission("corpora/sample-###/permissions/###");
console.log(res.getContentText());

In this case, no value is returned.

getPermissions

Get permission list of corpus. Ref

const res = CorporaApp.getPermissions("corpora/sample-###");
console.log(res);

"corpora/sample-###" is the resource name of corpus. When this script is run, the following result is obtained.

[
  {
    name: 'corpora/sample-###/permissions/###',
    granteeType: 'USER',
    emailAddress: '###email address###',
    role: 'OWNER'
  },
  ,
  ,
]

corpora/sample-###/permissions/### is the resource name of the permission.

updatePermission

Update permission of corpus. Ref

const res = CorporaApp.updatePermission(
  "corpora/sample-###/permissions/###",
  { role: "WRITER" },
  { updateMask: "role" }
);
console.log(res.getContentText());

corpora/sample-###/permissions/### is the resource name of permission. When this script is run, the following result is obtained.

{
  "name": "corpora/sample-###/permissions/###",
  "granteeType": "USER",
  "emailAddress": "###email address###",
  "role": "WRITER"
}

searchQueryWithGenerateAnswer

Search chunks using models.generateAnswer. Ref

const text = "###"; // Query
const source = "###"; // e.g. corpora/123 or corpora/123/documents/abc.

const requestBody = {
  contents: [{ parts: [{ text }], role: "user" }],
  answerStyle: "VERBOSE",
  semanticRetriever: { source, query: { parts: [{ text }] } }
};
const res = CorporaApp.searchQueryWithGenerateAnswer(requestBody);
console.log(res.getContentText());

Samples

Here, I would like to introduce sample scripts using this library.

Sample 1: Semantic search of my blog.

My blog can be seen at https://tanaikech.github.io/. In my blog, RSS can be seen. Ref This sample introduces a script for achieving the semantic search of my blog.

The flow of this sample is as follows.

1. Create a new corpus

If you have already had the corpus, please skip this.

function createCorpus() {
  const res = CorporaApp.createCorpus({ name: "corpora/sample-corpus", displayName: "sample corpus" });
  const { name } = JSON.parse(res.getContentText());
  console.log(name);
}

In the case of name: "corpora/sample-corpus",, the resource name of the corpus can be manually set as corpora/sample-corpus.

When this script is run, a new corpus is created. Please copy the value of name of the resource name of the created corpus.

2. Create a document in corpus

If you have already had the document, please skip this.

function createDocument() {
  const res = CorporaApp.createDocument(
    "corpora/sample-corpus", // Please set your resource name of created corpus.
    { name: "corpora/sample-corpus/documents/sample-document", displayName: "sample document" }
  );
  const { name } = JSON.parse(res.getContentText());
  console.log(name);
}

In the case of name: "corpora/sample-corpus/documents/sample-document",, the resource name of the document can be manually set as corpora/sample-corpus/documents/sample-document.

When this script is run, a new document is created in the corpus. Please copy the value of name of the resource name of the created document.

3. Put data to document

function setChunks() {
  const resourceNameOfdocument = "corpora/sample-corpus/documents/sample-document"; // Please set your resource name of document.

  const url = "https://tanaikech.github.io/post/index.xml"; // This is RSS of my blog (https://tanaikech.github.io/).
  const str = UrlFetchApp.fetch(url).getContentText();
  const xml = XmlService.parse(str);
  const root = xml.getRootElement();
  const ns = root.getNamespace();
  const items = root.getChild("channel", ns).getChildren("item");
  const keys = ["title", "link", "pubDate"];
  const requests = items.map(e => {
    const obj = new Map(keys.map(k => [k, e.getChild(k, ns).getValue()]));
    return {
      parent: resourceNameOfdocument,
      chunk: {
        data: { stringValue: obj.get("title") },
        customMetadata: [...obj].map(([key, stringValue]) => ({ key, stringValue }))
      }
    };
  });
  CorporaApp.setChunks(resourceNameOfdocument, { requests });
  console.log("Done.");
}

When this script is run, the blog data is put into the document. By this, the blog data can be searched as the semantic search.

When it is required to update the data in the document, in this case, I would like to recommend the following flow because of the process cost.

Delete the document.
Run the above function setChunk.

The sample script is as follows. In this case, it supposes that the corpus and document resource names are "corpora/sample-corpus" and "corpora/sample-corpus/documents/sample-document", respectively.

function updateChunk() {
  const corpusResourceName = "corpora/sample-corpus";
  const documentResourceName = "corpora/sample-corpus/documents/sample-document";

  CorporaApp.deleteDocument(documentResourceName, true);
  CorporaApp.createDocument(corpusResourceName, { name: documentResourceName, displayName: "sample document" });
  setChunks();
}

4. Testing

function searchQueryFromDocument() {
  const resourceNameOfdocument = "corpora/sample-corpus-###/documents/sample-document-###"; // Please set your resource name of document.
  const searchText = "Efficiently using Google Spreadsheets";

  const r = CorporaApp.searchQueryFromDocument(resourceNameOfdocument, { query: searchText, resultsCount: 3 });
  const { relevantChunks } = JSON.parse(r.getContentText());
  if (relevantChunks.length == 0) return;
  const res = relevantChunks.map(({ chunk: { customMetadata } }) => customMetadata);
  console.log(res);
}

When this script is run, the following result is obtained.

[
   [
      {
         "key":"title",
         "stringValue":"Report: Handling 10,000,000 cells in Google Spreadsheet using Google Apps Script"
      },
      {
         "key":"link",
         "stringValue":"https://tanaikech.github.io/2022/04/25/report-handling-10000000-cells-in-google-spreadsheet-using-google-apps-script/"
      },
      {
         "key":"pubDate",
         "stringValue":"Mon, 25 Apr 2022 15:06:49 +0900"
      }
   ],
   [
      {
         "key":"title",
         "stringValue":"Benchmark: Process Costs for Retrieving 1st Empty Cell and 1st Non Empty Cell of Specific Column in Google Spreadsheet using Google Apps Script"
      },
      {
         "key":"link",
         "stringValue":"https://tanaikech.github.io/2021/05/19/benchmark-process-costs-for-retrieving-1st-empty-cell-and-1st-non-empty-cell-of-specific-column-in-google-spreadsheet-using-google-apps-script/"
      },
      {
         "key":"pubDate",
         "stringValue":"Wed, 19 May 2021 13:47:29 +0900"
      }
   ],
   [
      {
         "key":"title",
         "stringValue":"Running Specific Function When Specific Sheet is Edited on Google Spreadsheet"
      },
      {
         "key":"link",
         "stringValue":"https://tanaikech.github.io/2020/10/04/running-specific-function-when-specific-sheet-is-edited-on-google-spreadsheet/"
      },
      {
         "key":"pubDate",
         "stringValue":"Sun, 04 Oct 2020 09:23:13 +0900"
      }
   ]
]

Sample 2: Semantic search of images.

When the images are searched with the semantic search, the following flow is run.

Create the descriptions of images using Gemini API.
Put the descriptions in the corpus.
Search images with the semantic search.

1. Create descriptions of images and put them as chunks to document

About 1 and 2 of the above flow, the script is as follows. Please copy and paste the following script. Please set the folder ID of the folder including images to folderId of the function setChunk.

In this script, it supposes that a document of corpora/sample-corpus/documents/sample-document has already been created in a corpus.

/**
 * ### Description
 * Generate text from text and image.
 * ref: https://medium.com/google-cloud/automatically-creating-descriptions-of-files-on-google-drive-using-gemini-pro-api-with-google-apps-7ef597a5b9fb
 *
 * @param {Object} object Object including API key, text, mimeType, and image data.
 * @return {String} Generated text.
 */
function getResFromImage_(object) {
  const { token, text, mime_type, data } = object;
  const url = `https://generativelanguage.googleapis.com/v1beta/models/gemini-pro-vision:generateContent`;
  const payload = { contents: [{ parts: [{ text }, { inline_data: { mime_type, data } }] }] };
  const options = {
    payload: JSON.stringify(payload),
    contentType: "application/json",
    headers: { authorization: "Bearer " + token }
  };
  const res = UrlFetchApp.fetch(url, options);
  const obj = JSON.parse(res.getContentText());
  if (obj.candidates.length > 0 && obj.candidates[0].content.parts.length > 0) {
    return obj.candidates[0].content.parts[0].text;
  }
  return "No response.";
}

function setChunk() {
  const documentResourceName = "corpora/sample-corpus/documents/sample-document"; // Please set the document resource name.
  const folderId = "###"; // Please set the folder ID of the folder including images.

  // 1. Retrieve description of the images using Gemini API.
  const requests = [];
  const files = DriveApp.getFolderById(folderId).searchFiles("trashed=false and mimeType contains 'image/'");
  const token = ScriptApp.getOAuthToken();
  while (files.hasNext()) {
    const file = files.next();
    const fileId = file.getId();
    const url = `https://drive.google.com/thumbnail?sz=w1000&id=${fileId}`;
    const bytes = UrlFetchApp.fetch(url, { headers: { authorization: "Bearer " + token } }).getContent();
    const base64 = Utilities.base64Encode(bytes);
    const description = getResFromImage_({ token, text: "What is this image? Explain within 50 words.", mime_type: "image/png", data: base64 });
    console.log(description);
    if (description == "No response.") continue;
    requests.push({
      parent: documentResourceName,
      chunk: {
        data: { stringValue: description.trim() },
        customMetadata: [{ key: "fileId", stringValue: fileId }, { key: "url", stringValue: file.getUrl() }]
      }
    });
  }
  if (requests.length == 0) return;

  // 2. Put descriptions to document as chunks.
  const res = CorporaApp.setChunks(documentResourceName, { requests });
  console.log(JSON.stringify(res.map(r => JSON.parse(r.getContentText()))));
}

When this script is run, the descriptions of the images in the folder are created by Gemini API. And, the description is put into a document in a corpus using Gemini API. As the next step, this document is used.

2. Testing

Please copy and paste the following script.

Please set your search text and the resource name of your document including the chunks with the above script.

function semanticSearch() {
  const searchText = "###"; // Please set your search text.
  const documentResourceName = "corpora/sample-corpus/documents/sample-document"; // Please set the document resource name.

  const res = CorporaApp.searchQueryFromDocument(documentResourceName, { query: searchText, resultsCount: 1 });
  const { relevantChunks } = JSON.parse(res.getContentText());
  if (!relevantChunks || relevantChunks.length == 0) return;
  const { data, customMetadata } = relevantChunks[0].chunk;
  const url = customMetadata.find(({ key }) => key == "url");
  console.log({ description: data.stringValue, url: url.stringValue });
}

When this script is run, the following result is obtained.

{
  description: "###",
  url: "https://drive.google.com/file/d/###/view?usp=drivesdk"
}

By this flow, the images on Google Drive can be searched with the semantic search.

Licence

MIT

Author

Tanaike

Donate

Update History

v1.0.0 (February 7, 2024)
1. Initial release.
v1.0.1 (February 16, 2024)
1. New method of searchQueryWithGenerateAnswer was added.
v1.0.2 (February 26, 2024)
1. New method of setAccessToken was added. When this method is used, you can use the access token retrieved from the service account. Default access token is retrieved by ScriptApp.getOAuthToken().
v1.0.3 (March 6, 2024)
1. New method of getChunk was added. When this method is used, you can retrieve a single chunk using the resource name of chunk.

TOP

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
images		images
CorporaApp.js		CorporaApp.js
LICENCE		LICENCE
README.md		README.md
appsscript.json		appsscript.json

License

tanaikech/CorporaApp

Folders and files

Latest commit

History

Repository files navigation

CorporaApp

Overview

Description

IMPORTANT

Library's project key

Usage

1. Install library

2. Linking Google Cloud Platform Project to Google Apps Script Project for New IDE

Scopes

Methods

setAccessToken

createCorpus

deleteCorpus

getCorpora

searchQueryFromCorpus

createDocument

deleteDocument

getDocuments

searchQueryFromDocument

setChunks

deleteChunks

getChunk

getChunks

updateChunks

createPermission

deletePermission

getPermissions

updatePermission

searchQueryWithGenerateAnswer

Samples

Sample 1: Semantic search of my blog.

1. Create a new corpus

2. Create a document in corpus

3. Put data to document

4. Testing

Sample 2: Semantic search of images.

1. Create descriptions of images and put them as chunks to document

2. Testing

Licence

Author

Update History

About

Topics

Resources

License

Stars

Watchers

Forks

Releases 4

Packages 0

Languages

Packages