@smythos/sre 1.7.18 → 1.7.40

package/src/Components/DataSourceIndexer.class.ts

@@ -8,6 +8,9 @@ import { SmythFS } from '@sre/IO/Storage.service/SmythFS.class';
 import { ConnectorService } from '@sre/Core/ConnectorsService';
 
 import { AccessCandidate } from '@sre/Security/AccessControl/AccessCandidate.class';
+import { TEmbeddings } from '@sre/IO/VectorDB.service/embed/BaseEmbedding';
+import { EmbeddingsFactory, SupportedModels } from '@sre/IO/VectorDB.service/embed';
+import { getLLMCredentials } from '@sre/LLMManager/LLM.service/LLMCredentials.helper';
 
 export class DataSourceIndexer extends Component {
     private MAX_ALLOWED_URLS_PER_INPUT = 20;
@@ -16,6 +19,9 @@ export class DataSourceIndexer extends Component {
         id: Joi.string().custom(validateCharacterSet, 'id custom validation').allow('').label('source identifier'),
         name: Joi.string().max(50).allow('').label('label'),
         metadata: Joi.string().allow(null).allow('').max(10000).label('metadata'),
+        chunkSize: Joi.number().optional(),
+        chunkOverlap: Joi.number().optional(),
+        version: Joi.string().valid('v1', 'v2').default('v1'),
     });
     constructor() {
         super();
@@ -23,6 +29,18 @@ export class DataSourceIndexer extends Component {
     init() {}
     async process(input, config, agent: Agent) {
         await super.process(input, config, agent);
+
+        let response: any = null;
+        if (!config.data.version || config.data.version === 'v1') {
+            response = await this.processV1(input, config, agent);
+        } else if (config.data.version === 'v2') {
+            response = await this.processV2(input, config, agent);
+        }
+
+        return response;
+    }
+
+    private async processV1(input, config, agent: Agent) {
         const teamId = agent.teamId;
         const agentId = agent.id;
         let debugOutput = agent.agentRuntime?.debug ? '== Source Indexer Log ==\n' : null;
@@ -72,47 +90,13 @@ export class DataSourceIndexer extends Component {
                 throw new Error(`Invalid id. Accepted characters: 'a-z', 'A-Z', '0-9', '-', '_', '.'`);
             }
 
-            // check if the datasource already exists with the same id
-            // await this.checkForRecordDuplicate(dsId, token);
-
             let indexRes: any = null;
             let parsedUrlArray: string[] | null = null;
-
-            //! DISABLE URL ARRAY PARSING FOR NOW UNTIL WE HAVE A GOOD WAY TO HANDLE BULK INDEXING
-            // if ((parsedUrlArray = parseUrlArray(inputSchema.value.Source))) {
-            //     debugOutput += `STEP: Parsing input as url array\n\n`;
-            //     if (parsedUrlArray.length > this.MAX_ALLOWED_URLS_PER_INPUT) {
-            //         throw new Error(`Too many urls in input. Max allowed: ${this.MAX_ALLOWED_URLS_PER_INPUT}`);
-            //     }
-
-            //     for (let url of parsedUrlArray) {
-            //         indexRes = await this.addDSFromUrl({
-            //             teamId,
-            //             namespaceId,
-            //             dsId, // WILL OVERRIDE EACH OTHER
-            //             type: detectURLSourceType(url),
-            //             url,
-            //             name: _config.name || 'Untitled',
-            //         });
-
-            //         debugOutput += `STEP: Created datasource for url: ${url}\n\n`;
-            //     }
-            // } else
-
             const dsId = DataSourceIndexer.genDsId(providedId, teamId, namespaceId);
 
             if (isUrl(inputSchema.value.Source)) {
                 debugOutput += `STEP: Parsing input as url\n\n`;
                 throw new Error('URLs are not supported yet');
-                // indexRes = await this.addDSFromUrl({
-                //     teamId,
-                //     namespaceId,
-                //     dsId,
-                //     type: detectURLSourceType(inputSchema.value.Source),
-                //     url: inputSchema.value.Source,
-                //     name: _config.name || 'Untitled',
-                //     metadata: _config.metadata || null,
-                // });
             } else {
                 debugOutput += `STEP: Parsing input as text\n\n`;
                 indexRes = await this.addDSFromText({
@@ -144,6 +128,136 @@ export class DataSourceIndexer extends Component {
         }
     }
 
+    private async processV2(input, config, agent: Agent) {
+        const teamId = agent.teamId;
+        const agentId = agent.id;
+        let debugOutput = agent.agentRuntime?.debug ? '== Source Indexer Log ==\n' : null;
+
+        try {
+            const _config = {
+                ...config.data,
+                name: TemplateString(config.data.name).parse(input).result,
+                id: TemplateString(config.data.id).parse(input).result,
+                metadata: TemplateString(config.data.metadata).parse(input).result,
+            };
+
+            const outputs = {};
+            for (let con of config.outputs) {
+                if (con.default) continue;
+                outputs[con.name] = con?.description ? `<${con?.description}>` : '';
+            }
+
+            // we try to get the namespace without the prefix teamId, if not exist, we use the full namespace id
+            const namespaceLabel = _config.namespace.split('_').slice(1).join('_') || _config.namespace;
+            const namespaceId = _config.namespace;
+            debugOutput += `[Selected namespace] \n${namespaceLabel}\n\n`;
+
+            // resolve the ns record, if not exist, throw an error (new in v2)
+            // then we also need to resolve the credentials
+            const nkvConnector = ConnectorService.getNKVConnector();
+            const nkvClient = nkvConnector.requester(AccessCandidate.team(teamId));
+            const rawNsRecord = await nkvClient.get(`vectorDB:namespaces`, namespaceId);
+
+            if (!rawNsRecord) {
+                return {
+                    _debug: debugOutput,
+                    _error: `Namespace ${namespaceLabel} does not exist`,
+                };
+            }
+
+            // const { credentialId, embeddings: embeddingsOptions } = JSON.parse(rawNsRecord.toString());
+            const namespaceRecord = JSON.parse(rawNsRecord.toString());
+            const accountConnector = ConnectorService.getAccountConnector();
+            const accountClient = accountConnector.requester(AccessCandidate.team(teamId));
+            const rawCredRecord = await accountClient.getTeamSetting(namespaceRecord.credentialId, 'vector_db_creds');
+            if (!rawCredRecord) {
+                throw new Error(`Credential ${namespaceRecord.credentialId} does not exist`);
+            }
+            const credRecord = JSON.parse(rawCredRecord);
+            await Promise.all(
+                Object.keys(credRecord.credentials).map(async (key) => {
+                    if (typeof credRecord.credentials[key] !== 'string') return;
+                    credRecord.credentials[key] = await TemplateString(credRecord.credentials[key]).parseTeamKeysAsync(teamId).asyncResult;
+                })
+            );
+
+            const vecDbConnector = ConnectorService.getVectorDBConnector(credRecord.provider).instance({
+                credentials: credRecord.credentials,
+                embeddings: await this.transformEmbedding(namespaceRecord.embeddings, config.data, teamId),
+            });
+            const vecDbClient = vecDbConnector.requester(AccessCandidate.team(teamId));
+
+            const inputSchema = this.validateInput(input);
+            if (inputSchema.error) {
+                throw new Error(`Input validation error: ${inputSchema.error}\n EXITING...`);
+            }
+
+            const providedId = _config.id;
+            // const isAutoId = _config.isAutoId;
+            const idRegex = /^[a-zA-Z0-9\-\_\.]+$/;
+
+            if (!providedId) {
+                // Assign a new ID if it's set to auto-generate or not provided
+                // _config.id = crypto.randomBytes(16).toString('hex');
+                throw new Error(`Id is required`);
+            } else if (!idRegex.test(providedId)) {
+                // Validate the provided ID if it's not auto-generated
+                throw new Error(`Invalid id. Accepted characters: 'a-z', 'A-Z', '0-9', '-', '_', '.'`);
+            }
+
+            const dsId = DataSourceIndexer.genDsId(providedId, teamId, namespaceLabel);
+
+            debugOutput += `STEP: Parsing input as text\n\n`;
+
+            const response = await vecDbClient.createDatasource(namespaceLabel, {
+                text: inputSchema.value.Source,
+                metadata: _config.metadata || null,
+                id: dsId,
+                label: _config.name || 'Untitled',
+            });
+
+            debugOutput += `Created datasource successfully\n\n`;
+
+            return {
+                _debug: debugOutput,
+                Success: {
+                    result: response || true,
+                    id: _config.id,
+                },
+                // _error,
+            };
+        } catch (err: any) {
+            debugOutput += `Error: ${err?.message || "Couldn't index data source"}\n\n`;
+            return {
+                _debug: debugOutput,
+                _error: err?.message || "Couldn't index data source",
+            };
+        }
+    }
+
+    private async transformEmbedding(embedding: { dimensions: string; modelId: string }, data: any, teamId: string): Promise<TEmbeddings> {
+        // we need to take this and return a proper TEmbeddings object
+
+        const provider = EmbeddingsFactory.getProviderByModel(embedding.modelId as any);
+
+        // based on the provider, we should be able to retreive the correct credentials
+        const modelsProvider = ConnectorService.getModelsProviderConnector();
+        const modelProviderCandidate = modelsProvider.requester(AccessCandidate.team(teamId));
+        const modelInfo = await modelProviderCandidate.getModelInfo(embedding.modelId);
+
+        const llmCreds = await getLLMCredentials(AccessCandidate.team(teamId), modelInfo);
+
+        return {
+            provider,
+            model: embedding.modelId,
+            credentials: llmCreds,
+            params: {
+                dimensions: parseInt(embedding.dimensions),
+                chunkSize: data.chunkSize,
+            },
+        };
+    }
+
     validateInput(input: any) {
         return Joi.object({
             Source: Joi.any().required(),

package/src/Components/GenAILLM.class.ts

@@ -528,26 +528,36 @@ export class GenAILLM extends Component {
             response = await contentPromise.catch((error) => {
                 return { error: error.message || error };
             });
-            // If the model stopped before completing the response, this is usually due to output token limit reached.
+
+            // #region Handle Response Errors
+            if (response?.error) {
+                const error = response?.error + ' ' + (response?.details || '');
+                logger.error(` LLM Error=`, error);
+
+                return { Output: response?.data, _error: error, _debug: logger.output };
+            }
+
+            const emptyResponseErrorMsg =
+                "Empty response. This is usually due to output token limit reached. Please try again with a higher 'Maximum Output Tokens'.";
+
+            // If the finish reason is not "stop", it means the model stopped before completing the response.
             if (finishReason !== 'stop') {
+                let errMsg = `The model stopped before completing the response.
+                \nReason: ${finishReason}.
+                \n${!response ? emptyResponseErrorMsg : ''}`;
+
                 return {
                     Reply: response,
-                    _error: 'The model stopped before completing the response, this is usually due to output token limit reached.',
+                    _error: errMsg,
                     _debug: logger.output,
                 };
             }
 
-            // in case we have the response but it's empty string, undefined or null
+            // If the finish reason is "stop" but there is still no response, it is usually caused by reaching the output token limit.
             if (!response) {
-                return { _error: ' LLM Error = Empty Response!', _debug: logger.output };
-            }
-
-            if (response?.error) {
-                const error = response?.error + ' ' + (response?.details || '');
-                logger.error(` LLM Error=`, error);
-
-                return { Output: response?.data, _error: error, _debug: logger.output };
+                return { _error: emptyResponseErrorMsg, _debug: logger.output };
             }
+            // #endregion
 
             const Reply = llmInference.connector.postProcess(response);
             if (Reply.error) {

package/src/Components/RAG/DataSourceCleaner.class.ts

@@ -0,0 +1,178 @@
+import { IAgent as Agent } from '@sre/types/Agent.types';
+import { Component } from '../Component.class';
+import { TemplateString } from '@sre/helpers/TemplateString.helper';
+import Joi from 'joi';
+import { validateCharacterSet } from '../../utils';
+import { ConnectorService } from '@sre/Core/ConnectorsService';
+import { AccessCandidate } from '@sre/Security/AccessControl/AccessCandidate.class';
+
+import { DataSourceIndexer } from './DataSourceIndexer.class';
+import { DataSourceComponent } from './DataSourceComponent.class';
+import { VectorDBConnector } from '@sre/IO/VectorDB.service/VectorDBConnector';
+
+export class DataSourceCleaner extends DataSourceComponent {
+    protected configSchema = Joi.object({
+        namespaceId: Joi.string().max(50).allow('').label('namespace'),
+        id: Joi.string().custom(validateCharacterSet, 'custom validation characterSet').allow('').label('source identifier'),
+        version: Joi.string().valid('v1', 'v2').default('v1'),
+    });
+    constructor() {
+        super();
+    }
+    init() {}
+
+    async process(input, config, agent: Agent) {
+        await super.process(input, config, agent);
+        if (!config.data.version || config.data.version === 'v1') {
+            return await this.processV1(input, config, agent);
+        } else if (config.data.version === 'v2') {
+            return await this.processV2(input, config, agent);
+        }
+    }
+    async processV1(input, config, agent: Agent) {
+        const teamId = agent.teamId;
+        const agentId = agent.id;
+        let debugOutput = agent.agentRuntime?.debug ? '== Source Indexer Log ==\n' : null;
+
+        try {
+            const configSchema = this.validateConfigData(config.data);
+            if (configSchema.error) {
+                throw new Error(`Config data validation error: ${configSchema.error}\n EXITING...`);
+            }
+
+            const outputs = {};
+            for (let con of config.outputs) {
+                if (con.default) continue;
+                outputs[con.name] = con?.description ? `<${con?.description}>` : '';
+            }
+
+            const inputSchema = this.validateInput(input);
+            if (inputSchema.error) {
+                throw new Error(`Input validation error: ${inputSchema.error}\n EXITING...`);
+            }
+
+            // const namespaceId = configSchema.value.namespaceId.split('_')?.slice(1).join('_') || configSchema.value.namespaceId;
+            const namespaceId = /^c[a-z0-9]{24}.+$/.test(configSchema.value.namespaceId)
+                ? configSchema.value.namespaceId.split('_').slice(1).join('_')
+                : configSchema.value.namespaceId;
+
+            let vectorDbConnector = ConnectorService.getVectorDBConnector();
+
+            let existingnamespace = await vectorDbConnector.requester(AccessCandidate.team(teamId)).namespaceExists(namespaceId);
+            if (!existingnamespace) {
+                throw new Error(`Namespace ${namespaceId} does not exist`);
+            }
+
+            const providedId = TemplateString(config.data.id).parse(input).result;
+            const idRegex = /^[a-zA-Z0-9\-\_\.]+$/;
+            if (!idRegex.test(providedId)) {
+                throw new Error(`Invalid id. Accepted characters: 'a-z', 'A-Z', '0-9', '-', '_', '.'`);
+            }
+            debugOutput += `Searching for data source with id: ${providedId}\n`;
+
+            const dsId = DataSourceIndexer.normalizeDsId(providedId, teamId, namespaceId);
+
+            await vectorDbConnector.requester(AccessCandidate.team(teamId)).deleteDatasource(namespaceId, dsId);
+
+            debugOutput += `Deleted data source with id: ${providedId}\n`;
+
+            return {
+                _debug: debugOutput,
+                Success: true,
+                // _error,
+            };
+        } catch (err: any) {
+            debugOutput += `Failed to delete data source: \n Error: ${err?.message}\n`;
+
+            return {
+                _debug: debugOutput,
+                _error: err?.message || "Couldn't delete data source",
+            };
+        }
+    }
+
+    async processV2(input, config, agent: Agent) {
+        const teamId = agent.teamId;
+        const agentId = agent.id;
+        let debugOutput = agent.agentRuntime?.debug ? '== Source Indexer Log ==\n' : null;
+
+        try {
+            const configSchema = this.validateConfigData(config.data);
+            if (configSchema.error) {
+                throw new Error(`Config data validation error: ${configSchema.error}\n EXITING...`);
+            }
+
+            const outputs = {};
+            for (let con of config.outputs) {
+                if (con.default) continue;
+                outputs[con.name] = con?.description ? `<${con?.description}>` : '';
+            }
+
+            const inputSchema = this.validateInput(input);
+            if (inputSchema.error) {
+                throw new Error(`Input validation error: ${inputSchema.error}\n EXITING...`);
+            }
+
+            // const namespaceId = configSchema.value.namespaceId.split('_')?.slice(1).join('_') || configSchema.value.namespaceId;
+            const namespaceLabel = /^c[a-z0-9]{24}.+$/.test(configSchema.value.namespaceId)
+                ? configSchema.value.namespaceId.split('_').slice(1).join('_')
+                : configSchema.value.namespaceId;
+            const namespaceId = configSchema.value.namespaceId;
+
+            let vecDbConnector: VectorDBConnector = null;
+            try {
+                vecDbConnector = await this.resolveVectorDbConnector(namespaceId, teamId);
+            } catch (err: any) {
+                debugOutput += `Error: ${err?.message || "Couldn't get vector database connector"}\n\n`;
+                return {
+                    _debug: debugOutput,
+                    _error: err?.message || "Couldn't get vector database connector",
+                };
+            }
+
+            // let existingnamespace = await vecDbConnector.requester(AccessCandidate.team(teamId)).namespaceExists(namespaceId);
+            // if (!existingnamespace) {
+            //     throw new Error(`Namespace ${namespaceId} does not exist`);
+            // }
+
+            const providedId = TemplateString(config.data.id).parse(input).result;
+            const idRegex = /^[a-zA-Z0-9\-\_\.]+$/;
+            if (!idRegex.test(providedId)) {
+                throw new Error(`Invalid id. Accepted characters: 'a-z', 'A-Z', '0-9', '-', '_', '.'`);
+            }
+            debugOutput += `Searching for data source with id: ${providedId}\n`;
+
+            const dsId = DataSourceComponent.normalizeDsId(providedId, teamId, namespaceLabel);
+
+            await vecDbConnector.requester(AccessCandidate.team(teamId)).deleteDatasource(namespaceLabel, dsId);
+
+            debugOutput += `Deleted data source with id: ${providedId}\n`;
+
+            return {
+                _debug: debugOutput,
+                Success: true,
+                // _error,
+            };
+        } catch (err: any) {
+            debugOutput += `Failed to delete data source: \n Error: ${err?.message}\n`;
+
+            return {
+                _debug: debugOutput,
+                _error: err?.message || "Couldn't delete data source",
+            };
+        }
+    }
+
+    validateInput(input: any) {
+        return Joi.object({}).unknown(true).validate(input);
+    }
+
+    validateConfigData(data: any) {
+        return Joi.object({
+            namespaceId: Joi.string().required(),
+            id: Joi.string().optional().allow('').allow(null),
+        })
+            .unknown(true)
+            .validate(data);
+    }
+}

package/src/Components/RAG/DataSourceComponent.class.ts

@@ -0,0 +1,111 @@
+// Base class for the RAG components that handle the shared logic for the RAG components
+import { EmbeddingsFactory } from '@sre/IO/VectorDB.service/embed';
+import { Component } from '../Component.class';
+import { ConnectorService } from '@sre/Core/ConnectorsService';
+import { AccessCandidate } from '@sre/Security/AccessControl/AccessCandidate.class';
+import { getLLMCredentials } from '@sre/LLMManager/LLM.service/LLMCredentials.helper';
+import { TEmbeddings } from '@sre/IO/VectorDB.service/embed/BaseEmbedding';
+import { TemplateString } from '@sre/helpers/TemplateString.helper';
+import { VectorDBConnector } from '@sre/IO/VectorDB.service/VectorDBConnector';
+import { TLLMCredentials } from '@sre/types/LLM.types';
+
+export type NsRecord = {
+    credentialId: string;
+    embeddings: { dimensions: string; modelId: string };
+    label: string;
+    createdAt: string;
+};
+
+export enum TDataSourceCompErrorCodes {
+    NAMESPACE_NOT_FOUND = 1,
+    CREDENTIAL_NOT_FOUND = 2,
+    EMBEDDING_CONFIG_ERROR = 3,
+}
+export class DataSourceCompError extends Error {
+    public code: TDataSourceCompErrorCodes;
+    constructor(message: string, code: TDataSourceCompErrorCodes) {
+        super(message);
+        this.name = 'DataSourceCompError';
+        this.code = code;
+    }
+}
+export class DataSourceComponent extends Component {
+    constructor() {
+        super();
+    }
+
+    public async resolveVectorDbConnector(namespace: string | NsRecord, teamId: string): Promise<VectorDBConnector> {
+        // resolve the ns record, if not exist, throw an error (new in v2)
+        // then we also need to resolve the credentials
+        let namespaceRecord = namespace as NsRecord;
+
+        if (typeof namespace === 'string') {
+            // if it's a string, we need to get the namespace record from the NKV
+            const nkvConnector = ConnectorService.getNKVConnector();
+            const nkvClient = nkvConnector.requester(AccessCandidate.team(teamId));
+            const rawNsRecord = await nkvClient.get(`vectorDB:namespaces`, namespace);
+
+            if (!rawNsRecord) {
+                throw new DataSourceCompError(`Namespace ${namespace} does not exist`, TDataSourceCompErrorCodes.NAMESPACE_NOT_FOUND);
+            }
+
+            // const { credentialId, embeddings: embeddingsOptions } = JSON.parse(rawNsRecord.toString());
+            namespaceRecord = JSON.parse(rawNsRecord.toString()) as NsRecord;
+        }
+
+        const accountConnector = ConnectorService.getAccountConnector();
+        const accountClient = accountConnector.requester(AccessCandidate.team(teamId));
+        const rawCredRecord = await accountClient.getTeamSetting(namespaceRecord.credentialId, 'vector_db_creds');
+        if (!rawCredRecord) {
+            throw new DataSourceCompError(
+                `Credential ${namespaceRecord.credentialId} does not exist`,
+                TDataSourceCompErrorCodes.CREDENTIAL_NOT_FOUND
+            );
+        }
+        const credRecord = JSON.parse(rawCredRecord);
+        await Promise.all(
+            Object.keys(credRecord.credentials).map(async (key) => {
+                if (typeof credRecord.credentials[key] !== 'string') return;
+                credRecord.credentials[key] = await TemplateString(credRecord.credentials[key]).parseTeamKeysAsync(teamId).asyncResult;
+            })
+        );
+
+        const vecDbConnector = ConnectorService.getVectorDBConnector(credRecord.provider).instance({
+            credentials: credRecord.credentials,
+            embeddings: await this.buildEmbeddingConfig(namespaceRecord.embeddings, teamId),
+        });
+
+        return vecDbConnector;
+    }
+
+    public async buildEmbeddingConfig(embedding: { dimensions: string; modelId: string }, teamId: string): Promise<TEmbeddings> {
+        // we need to take this and return a proper TEmbeddings object
+
+        const provider = EmbeddingsFactory.getProviderByModel(embedding.modelId as any);
+
+        // based on the provider, we should be able to retreive the correct credentials
+        const modelsProvider = ConnectorService.getModelsProviderConnector();
+        const modelProviderCandidate = modelsProvider.requester(AccessCandidate.team(teamId));
+        // const modelInfo = await modelProviderCandidate.getModelInfo(embedding.modelId);
+
+        const llmCreds = await getLLMCredentials(AccessCandidate.team(teamId), {
+            provider,
+            modelId: embedding.modelId,
+            credentials: [TLLMCredentials.Vault, TLLMCredentials.Internal],
+        });
+
+        return {
+            provider,
+            model: embedding.modelId,
+            credentials: llmCreds,
+            dimensions: parseInt(embedding.dimensions), // pass both for backwards compatibility
+            params: {
+                dimensions: parseInt(embedding.dimensions), // pass both for backwards compatibility
+            },
+        };
+    }
+
+    public static normalizeDsId(providedId: string, teamId: string, namespaceId: string) {
+        return `${teamId}::${namespaceId}::${providedId}`;
+    }
+}