@tmlmobilidade/ai 20260504.1029.41

package/README.md

@@ -0,0 +1,3 @@
+## @tmlmobilidade/ai
+
+This package provides a wrapper for OCI Generative AI services.

package/dist/index.d.ts

@@ -0,0 +1 @@
+export * from './oci-generativeai.js';

package/dist/index.js

@@ -0,0 +1 @@
+export * from './oci-generativeai.js';

package/dist/oci-generativeai.d.ts

@@ -0,0 +1,19 @@
+/**
+ * This is a simple wrapper around the OCI Generative AI SDK to make it easier
+ * to use in our application. It currently only supports running a text prompt
+ * through the "openai.gpt-oss-120b" model and getting a response.
+ */
+export declare class OCIGenerativeAIProvider {
+    private readonly modelOCID;
+    private readonly ociClient;
+    constructor();
+    /**
+     * Runs a prompt through the OCI Generative AI service and returns the response.
+     * This is using the "openai.gpt-oss-120b" model, which is a 120B parameter model
+     * based on the open source Falcon-40B-Instruct-v2 architecture,
+     * but with more parameters and trained on more data.
+     * @param prompt The prompt to run through the model.
+     * @returns The response from the model.
+     */
+    run(prompt: string): Promise<string>;
+}

package/dist/oci-generativeai.js

@@ -0,0 +1,95 @@
+/* * */
+import { readFileSync } from 'node:fs';
+import { NoRetryConfigurationDetails, Region, SimpleAuthenticationDetailsProvider } from 'oci-common';
+import { GenerativeAiInferenceClient } from 'oci-generativeaiinference';
+/**
+ * This is a simple wrapper around the OCI Generative AI SDK to make it easier
+ * to use in our application. It currently only supports running a text prompt
+ * through the "openai.gpt-oss-120b" model and getting a response.
+ */
+export class OCIGenerativeAIProvider {
+    //
+    modelOCID = 'ocid1.generativeaimodel.oc1.eu-frankfurt-1.amaaaaaask7dceyaz25iqdyw27r4kve6tu7mnbpviyllkgswzwkwstriguvq';
+    ociClient;
+    constructor() {
+        // Validate that all required environment variables are set
+        if (!process.env.OCI_FINGERPRINT)
+            throw new Error('Missing required environment variables for OCI Generative AI Provider');
+        if (!process.env.OCI_PRIVATE_KEY_PATH && !process.env.OCI_PRIVATE_KEY)
+            throw new Error('Missing required environment variables for OCI Generative AI Provider: either OCI_PRIVATE_KEY_PATH or OCI_PRIVATE_KEY must be set');
+        if (!process.env.OCI_REGION)
+            throw new Error('Missing required environment variables for OCI Generative AI Provider');
+        if (!process.env.OCI_TENANCY)
+            throw new Error('Missing required environment variables for OCI Generative AI Provider');
+        if (!process.env.OCI_USER)
+            throw new Error('Missing required environment variables for OCI Generative AI Provider');
+        if (!process.env.OCI_COMPARTMENT)
+            throw new Error('Missing required environment variables for OCI Generative AI Provider');
+        // Build the OCI client using the environment variables for authentication
+        this.ociClient = new GenerativeAiInferenceClient({
+            authenticationDetailsProvider: new SimpleAuthenticationDetailsProvider(process.env.OCI_TENANCY, process.env.OCI_USER, process.env.OCI_FINGERPRINT, process.env.OCI_PRIVATE_KEY_PATH ? readFileSync(process.env.OCI_PRIVATE_KEY_PATH, 'utf8') : process.env.OCI_PRIVATE_KEY, null, Region.fromRegionId(process.env.OCI_REGION)),
+        });
+        this.ociClient.endpoint = `https://inference.generativeai.${process.env.OCI_REGION}.oci.oraclecloud.com`;
+    }
+    /**
+     * Runs a prompt through the OCI Generative AI service and returns the response.
+     * This is using the "openai.gpt-oss-120b" model, which is a 120B parameter model
+     * based on the open source Falcon-40B-Instruct-v2 architecture,
+     * but with more parameters and trained on more data.
+     * @param prompt The prompt to run through the model.
+     * @returns The response from the model.
+     */
+    async run(prompt) {
+        //
+        //
+        // Build the chat request, in the most complicated way possible
+        // because the OCI SDK is a nightmare to work with.
+        // (This was written by AI, I swear.)
+        const chatRequest = {
+            chatDetails: {
+                chatRequest: {
+                    apiFormat: 'GENERIC',
+                    frequencyPenalty: 0,
+                    maxTokens: 2048,
+                    messages: [
+                        {
+                            content: [
+                                {
+                                    // @ts-expect-error — OMG this has to be the most convoluted way to send
+                                    // a simple text message I've ever seen in my life.
+                                    text: prompt,
+                                    type: 'TEXT',
+                                },
+                            ],
+                            role: 'USER',
+                        },
+                    ],
+                    presencePenalty: 0,
+                    temperature: 1,
+                    topK: 0,
+                    topP: 1,
+                },
+                compartmentId: process.env.OCI_COMPARTMENT,
+                servingMode: {
+                    modelId: this.modelOCID,
+                    servingType: 'ON_DEMAND',
+                },
+            },
+            retryConfiguration: NoRetryConfigurationDetails,
+        };
+        //
+        // Send the chat request and return the response.
+        // Again, this is the most convoluted way to get a simple text response from the model.
+        const chatResponse = await this.ociClient.chat(chatRequest);
+        if (!('chatResult' in chatResponse)) {
+            throw new Error('Invalid response from OCI Generative AI service: ' + JSON.stringify(chatResponse));
+        }
+        if (!('choices' in chatResponse.chatResult.chatResponse)) {
+            throw new Error('Invalid response from OCI Generative AI service: ' + JSON.stringify(chatResponse));
+        }
+        // @ts-expect-error — Yes, I know this is a nightmare to read,
+        // but this is the documented way to get the text response from the model.
+        return chatResponse.chatResult.chatResponse.choices[0].message.content[0].text;
+        //
+    }
+}

package/package.json

@@ -0,0 +1,49 @@
+{
+	"name": "@tmlmobilidade/ai",
+	"version": "20260504.1029.41",
+	"author": {
+		"email": "iso@tmlmobilidade.pt",
+		"name": "TML-ISO"
+	},
+	"license": "AGPL-3.0-or-later",
+	"homepage": "https://github.com/tmlmobilidade/go#readme",
+	"bugs": {
+		"url": "https://github.com/tmlmobilidade/go/issues"
+	},
+	"repository": {
+		"type": "git",
+		"url": "git+https://github.com/tmlmobilidade/go.git"
+	},
+	"keywords": [
+		"public transit",
+		"tml",
+		"transportes metropolitanos de lisboa",
+		"services"
+	],
+	"publishConfig": {
+		"access": "public"
+	},
+	"type": "module",
+	"files": [
+		"dist"
+	],
+	"main": "./dist/index.js",
+	"types": "./dist/index.d.ts",
+	"scripts": {
+		"build": "tsc && resolve-tspaths",
+		"lint": "eslint && tsc --noEmit",
+		"lint:fix": "eslint --fix",
+		"watch": "tsc-watch --onSuccess 'resolve-tspaths'"
+	},
+	"dependencies": {
+		"oci-common": "2.131.0",
+		"oci-generativeaiinference": "2.131.0"
+	},
+	"devDependencies": {
+		"@tmlmobilidade/tsconfig": "*",
+		"@types/node": "25.6.0",
+		"resolve-tspaths": "0.8.23",
+		"tsc-watch": "7.2.0",
+		"typescript": "5.9.3"
+	}
+}