@jutge.org/toolkit 4.2.30 → 4.2.34

package/docs/install-linux.md

@@ -66,33 +66,6 @@ Depending on your needs, you may want to install:
     sudo apt-get install build-essential
     ```
 
-## Setting Up AI Features (Optional)
-
-If you want to use JutgeAI features to generate problems and content, you need to set up API keys:
-
-### For Google Gemini (Free for UPC users):
-
-1. Visit https://aistudio.google.com/ and sign in
-2. Click "Get API key" in the sidebar
-3. Click "Create API key"
-4. Copy the generated key
-5. Add to `~/.bashrc` or `~/.zshrc`:
-
-    ```bash
-    export GEMINI_API_KEY="your-key-here"
-    ```
-
-### For OpenAI (Paid):
-
-1. Create an account at https://platform.openai.com/
-2. Navigate to API Keys section
-3. Create a new secret key
-4. Add to `~/.bashrc` or `~/.zshrc`:
-
-    ```bash
-    export OPENAI_API_KEY="your-key-here"
-    ```
-
 ## Troubleshooting
 
 **Command not found after installation:**

package/docs/install-macos.md

@@ -70,33 +70,6 @@ Then install the tools you need:
     xcode-select --install
     ```
 
-## Setting Up AI Features (Optional)
-
-If you want to use JutgeAI features to generate problems and content, you need to set up API keys:
-
-### For Google Gemini (Free for UPC users):
-
-1. Visit https://aistudio.google.com/ and sign in
-2. Click "Get API key" in the sidebar
-3. Click "Create API key"
-4. Copy the generated key
-5. Add to `~/.bashrc` or `~/.zshrc`:
-
-    ```bash
-    export GEMINI_API_KEY="your-key-here"
-    ```
-
-### For OpenAI (Paid):
-
-1. Create an account at https://platform.openai.com/
-2. Navigate to API Keys section
-3. Create a new secret key
-4. Add to `~/.bashrc` or `~/.zshrc`:
-
-    ```bash
-    export OPENAI_API_KEY="your-key-here"
-    ```
-
 ## Troubleshooting
 
 **Command not found after installation:**

package/docs/install-windows.md

@@ -57,33 +57,6 @@ This command will show which tools are installed on your system.
     2. Extract to a folder (e.g., `C:\w64devkit`)
     3. Run `w64devkit.exe` to open a terminal with GCC available
 
-## Setting Up AI Features (Optional)
-
-If you want to use JutgeAI features to generate problems and content, you need to set up API keys:
-
-### For Google Gemini (Free for UPC users):
-
-1. Visit https://aistudio.google.com/ and sign in
-2. Click "Get API key" in the sidebar
-3. Click "Create API key"
-4. Copy the generated key
-5. Set the environment variable permanently:
-
-    ```powershell
-    [System.Environment]::SetEnvironmentVariable('GEMINI_API_KEY', 'your-key-here', 'User')
-    ```
-
-### For OpenAI (Paid):
-
-1. Create an account at https://platform.openai.com/
-2. Navigate to API Keys section
-3. Create a new secret key
-4. Set the environment variable permanently:
-
-    ```powershell
-    [System.Environment]::SetEnvironmentVariable('OPENAI_API_KEY', 'your-key-here', 'User')
-    ```
-
 ## Troubleshooting
 
 **Command not found after installation:**

package/docs/jutge-ai.md

@@ -29,54 +29,42 @@ In particular, Jutge<sup>AI</sup> features can assist in:
 
     - You can create new test case generators to extend the existing test suite.
 
-    - You can generate `award.png` and `award.html` files for the problem. (Note: `award.png` requires `dall-e-3` model access.)
+    - You can generate `award.png` and `award.html` files for the problem.
 
 As in any other use of AI and LLMs, it is important to review and validate the generated content to ensure its correctness and quality. Treat the generated content as a first draft that needs to be refined and validated.
 
-In order to use the Jutge<sup>AI</sup> features of the toolkit, you need to have API keys for the models you wish to use. You should get the keys from the respective providers and set them as environment variables in your system. Because of the costs associated to the use of these models, the toolkit or Jutge.org cannot provide these keys directly.
+# Jutge<sup>AI</sup> models
 
-UPC users can get free access to Gemini models through their institutional Google accounts. These work well with Jutge<sup>AI</sup> features but have usage limits. If you need more capacity, consider using an OpenAI API key.
+The toolkit currently supports the following models:
 
-## How to get Gemini API key
+### Google Gemini
 
-1. Visit Google AI Studio:
+Google Gemini is fast and free for UPC users but its rate limits are so low it is almost impossible to use it for practical purposes.
 
-    Go to [aistudio.google.com](https://aistudio.google.com/) and sign in with your Google Account.
+Available models:
 
-2. Access the API Section:
+- google/gemini-2.5-flash
+- google/gemini-2.5-flash-lite
 
-    Click on the **Get API key** button located in the left-hand sidebar menu.
+### OpenAI GPT
 
-3. Generate the Key:
-    - Click the **Create API key** button.
-    - You will have two options: **Create API key in new project** (recommended for beginners) or **Create API key in existing project.**
-    - Select your preference, and the system will generate a unique key for you.
+OpenAI is slower but more reliable and has a higher rate limit. However, it is not free and requires a paid account.
 
-4. Secure Your Key:
+Available models:
 
-    Copy the generated key immediately.
+- openai/gpt-5-nano
+- openai/gpt-5-mini
+- openai/gpt-4.1-nano
+- openai/gpt-4.1-mini
 
-5. Set `GEMINI_API_KEY` environment variable with the obtained key to use it in the toolkit. This will make models such as `google/gemini-2.5-flash` or `google/gemini-2.5-flash-lite` available for Jutge<sup>AI</sup> features.
+See https://platform.openai.com/docs/pricing for the pricing of the OpenAI models.
 
-## How to get OpenAI API key
+### Recommendation
 
-1. Create an OpenAI Account:
+Try to use `gpt-4.1-nano` or `gpt-4.1-mini` for the quickest results. If you need more reliable results, use `gpt-5-nano` or `gpt-5-mini`.
 
-    Go to the OpenAI website and sign up (or log in if you already have an account).
+# Jutge<sup>AI</sup> costs
 
-2. Access the API Dashboard:
+In order to use the Jutge<sup>AI</sup> features of the toolkit, we have allocated a small budget to cover the costs associated to pay for the use of the models. This budget is shared by all instructors of Jutge.org. Jutge.org records the costs incurred for each instructor as estimated from input and output token counts. Rate limits are applied to avoid abuse or misuse.
 
-    After logging in, open the **API dashboard** from your account menu.
-
-3. Create an API Key:
-    - Navigate to **API Keys**.
-    - Click **Create new secret key**.
-    - Copy the key immediately (it will not be shown again).
-
-4. Secure Your Key: Copy the generated key immediately.
-
-5. Set `OPENAI_API_KEY` environment variable with the obtained key to use it in the toolkit.This will make models such as `openai/gpt-5-mini` or `openai/gpt-5-nano` available for Jutge<sup>AI</sup> features.
-
-## Other models
-
-We use `multi-llm-ts` package to interface with different models. If you have access to other models supported by this package, you can set the corresponding environment variables as described in the [multi-llm-ts documentation](https://github.com/nbonamy/multi-llm-ts) to use them with Jutge<sup>AI</sup> features.
+Please contact the Jutge.org team if you need to increase your budget.

package/package.json

@@ -1,7 +1,7 @@
 {
     "name": "@jutge.org/toolkit",
     "description": "Toolkit to prepare problems for Jutge.org",
-    "version": "4.2.30",
+    "version": "4.2.34",
     "homepage": "https://jutge.org",
     "author": {
         "name": "Jutge.org",
@@ -79,16 +79,12 @@
         "handlebars": "^4.7.8",
         "image-size": "^2.0.2",
         "inquirer-checkbox-plus-plus": "^1.1.1",
-        "multi-llm-ts": "^5.0.1",
         "nanoid": "^5.1.6",
         "open": "^11.0.0",
-        "openai": "^6.17.0",
-        "ora": "^9.1.0",
         "pretty-bytes": "^7.1.0",
         "pretty-ms": "^9.3.0",
         "radash": "^12.1.1",
         "semver": "^7.7.3",
-        "sharp": "^0.34.5",
         "terminal-link": "^5.0.0",
         "tree-node-cli": "^1.6.0",
         "yaml": "^2.8.2",
@@ -97,13 +93,12 @@
         "zod-validation-error": "^5.0.0"
     },
     "devDependencies": {
-        "prettier": "^3.8.1",
-        "typescript-eslint": "^8.54.0",
         "@types/archiver": "^7.0.0",
         "@types/image-size": "^0.8.0",
         "@types/node": "^25.1.0",
-        "@types/ora": "^3.2.0",
-        "@types/semver": "^7.7.1"
+        "@types/semver": "^7.7.1",
+        "prettier": "^3.8.1",
+        "typescript-eslint": "^8.54.0"
     },
     "peerDependencies": {
         "typescript": "^5.9.3"

package/toolkit/ask.ts

@@ -1,11 +1,12 @@
 import { Command } from '@commander-js/extra-typings'
 import { glob } from 'fs/promises'
-import { complete } from '../lib/ai'
+import { complete } from '../lib/aiclient'
 import { settings } from '../lib/settings'
 import tui from '../lib/tui'
 import { projectDir, readTextInDir } from '../lib/utils'
 import { join } from 'path'
 import { all } from 'radash'
+import { getLoggedInJutgeClient } from '../lib/login'
 
 export const askCmd = new Command('ask')
     .description('Ask questions about jutge-toolkit to JutgeAI')
@@ -14,12 +15,13 @@ export const askCmd = new Command('ask')
     .option('-m, --model <model>', 'AI model to use', settings.defaultModel)
 
     .action(async (question, { model }) => {
+        const jutge = await getLoggedInJutgeClient()
         const systemPrompt = await readTextInDir(join(projectDir(), 'assets', 'prompts', 'ask'), 'ask.md')
 
         const docs = await loadDocumentation() // Load your markdown files
         const fullPrompt = `${docs}\n\nUser question: ${question}`
 
-        const answer = await complete(model, systemPrompt, fullPrompt)
+        const answer = await complete(jutge, model, systemPrompt, fullPrompt)
         await tui.markdown(answer)
         tui.warning(
             `This answer generated by JutgeAI using ${model} is not authoritative but we hope it will help you.`,

package/toolkit/generate.ts

@@ -1,8 +1,6 @@
 import { Argument, Command } from '@commander-js/extra-typings'
-import { join } from 'path'
-import sharp from 'sharp'
-import { complete, generateImage } from '../lib/ai'
-import { languageKeys, languageNames, proglangKeys } from '../lib/data'
+import { createProblemWithJutgeAI } from '../lib/create-with-jutgeai'
+import { languageKeys, languageNames, proglangKeys, proglangNames } from '../lib/data'
 import {
     addAlternativeSolution,
     addMainFile,
@@ -12,8 +10,7 @@ import {
 import { newProblem } from '../lib/problem'
 import { settings } from '../lib/settings'
 import tui from '../lib/tui'
-import { writeText } from '../lib/utils'
-import { createProblemWithJutgeAI } from '../lib/create-with-jutgeai'
+import { getLoggedInJutgeClient } from '../lib/login'
 
 export const generateCmd = new Command('generate')
     .description('Generate problem elements using JutgeAI')
@@ -33,7 +30,10 @@ generateCmd
     .option('-m, --model <model>', 'AI model to use', settings.defaultModel)
 
     .action(async ({ input, output, directory, model, doNotAsk }) => {
-        await createProblemWithJutgeAI(model, directory, input, output, doNotAsk)
+        const jutge = await getLoggedInJutgeClient()
+        await tui.section('Generating problem with JutgeAI', async () => {
+            await createProblemWithJutgeAI(jutge, model, directory, input, output, doNotAsk)
+        })
     })
 
 generateCmd
@@ -47,8 +47,8 @@ The original statement will be used as the source text for translation.
 
 Provide one or more target language from the following list:
 ${Object.entries(languageNames)
-    .map(([key, name]) => `  - ${key}: ${name}`)
-    .join('\n')}
+            .map(([key, name]) => `  - ${key}: ${name}`)
+            .join('\n')}
 
 The added translations will be saved in the problem directory overwrite possible existing files.`,
     )
@@ -58,10 +58,11 @@ The added translations will be saved in the problem directory overwrite possible
     .option('-m, --model <model>', 'AI model to use', settings.defaultModel)
 
     .action(async (languages, { directory, model }) => {
+        const jutge = await getLoggedInJutgeClient()
         const problem = await newProblem(directory)
         await tui.section('Generating statement translations', async () => {
             for (const language of languages) {
-                await addStatementTranslation(model, problem, language)
+                await addStatementTranslation(jutge, model, problem, language)
             }
         })
     })
@@ -77,8 +78,8 @@ The golden solution will be used as a reference for generating the alternatives.
 
 Provide one or more target programming languages from the following list:
 ${Object.entries(languageNames)
-    .map(([key, name]) => `  - ${key}: ${name}`)
-    .join('\n')}
+            .map(([key, name]) => `  - ${key}: ${name}`)
+            .join('\n')}
 
 The added solutions will be saved in the problem directory overwrite possible existing files.`,
     )
@@ -88,10 +89,15 @@ The added solutions will be saved in the problem directory overwrite possible ex
     .option('-m, --model <model>', 'AI model to use', settings.defaultModel)
 
     .action(async (proglangs, { directory, model }) => {
+        const jutge = await getLoggedInJutgeClient()
         const problem = await newProblem(directory)
-        for (const proglang of proglangs) {
-            await addAlternativeSolution(model, problem, proglang)
-        }
+        await tui.section('Generating statement translations', async () => {
+            for (const proglang of proglangs) {
+                await tui.section(`Generating solution in ${proglangNames[proglang]}`, async () => {
+                    await addAlternativeSolution(jutge, model, problem, proglang)
+                })
+            }
+        })
     })
 
 generateCmd
@@ -107,8 +113,8 @@ The main file for the golden solution will be used as a reference for generating
 
 Provide one or more target programming languages from the following list:
 ${Object.entries(languageNames)
-    .map(([key, name]) => `  - ${key}: ${name}`)
-    .join('\n')}
+            .map(([key, name]) => `  - ${key}: ${name}`)
+            .join('\n')}
 
 The added main files will be saved in the problem directory overwrite possible existing files.`,
     )
@@ -118,9 +124,10 @@ The added main files will be saved in the problem directory overwrite possible e
     .option('-m, --model <model>', 'AI model to use', settings.defaultModel)
 
     .action(async (proglangs, { directory, model }) => {
+        const jutge = await getLoggedInJutgeClient()
         const problem = await newProblem(directory)
         for (const proglang of proglangs) {
-            await addMainFile(model, problem, proglang)
+            await addMainFile(jutge, model, problem, proglang)
         }
     })
 
@@ -138,14 +145,15 @@ generateCmd
     .option('-m, --model <model>', 'AI model to use', settings.defaultModel)
 
     .action(async ({ efficiency, hard, random, all, directory, model, output }) => {
+        const jutge = await getLoggedInJutgeClient()
         const problem = await newProblem(directory)
         await tui.section('Generating test cases generators', async () => {
-            if (all || random) await generateTestCasesGenerator(model, problem, output, 'random')
-            if (all || hard) await generateTestCasesGenerator(model, problem, output, 'hard')
-            if (all || efficiency) await generateTestCasesGenerator(model, problem, output, 'efficiency')
+            if (all || random) await generateTestCasesGenerator(jutge, model, problem, output, 'random')
+            if (all || hard) await generateTestCasesGenerator(jutge, model, problem, output, 'hard')
+            if (all || efficiency) await generateTestCasesGenerator(jutge, model, problem, output, 'efficiency')
         })
     })
-
+/*
 generateCmd
     .command('award.png')
     .summary('Generate award.png using JutgeAI')
@@ -216,3 +224,4 @@ The new message will be saved as award.html in the problem directory, overriding
         await writeText(output, message)
         tui.success(`Added ${output}`)
     })
+*/

package/toolkit/index.ts

@@ -6,7 +6,6 @@ import { fromError } from 'zod-validation-error'
 import { settings } from '../lib/settings'
 import { packageJson } from '../lib/versions'
 import { aboutCmd } from './about'
-import { aiCmd } from './ai'
 import { verifyCmd } from './verify'
 import { cleanCmd } from './clean'
 import { compilersCmd } from './compilers'
@@ -43,7 +42,6 @@ program.addCommand(doctorCmd)
 if (settings.developer) {
     program.addCommand(quizCmd)
     program.addCommand(compilersCmd)
-    program.addCommand(aiCmd)
 }
 program.addCommand(configCmd)
 program.addCommand(upgradeCmd)

package/toolkit/ai.ts

@@ -1,56 +0,0 @@
-import { Command } from '@commander-js/extra-typings'
-import sharp from 'sharp'
-import z from 'zod'
-import { complete, generateImage, listModels } from '../lib/ai.ts'
-import { settings } from '../lib/settings.ts'
-import tui from '../lib/tui.ts'
-import { convertStringToItsType } from '../lib/utils.ts'
-
-export const aiCmd = new Command('ai')
-    .description('Query AI models')
-
-    .action(() => {
-        aiCmd.help()
-    })
-
-aiCmd
-    .command('models')
-    .description('Show available AI models')
-
-    .action(async () => {
-        const models = await listModels()
-        tui.yaml(models)
-    })
-
-aiCmd
-    .command('complete')
-    .description('Complete a prompt using an AI model')
-
-    .argument('<prompt>', 'the user prompt to complete')
-    .option('-s, --system-prompt <system>', 'the system prompt to use', 'You are a helpful assistant.')
-    .option('-m, --model <model>', 'the AI model to use', settings.defaultModel)
-
-    .action(async (prompt, { model, systemPrompt }) => {
-        prompt = prompt.trim()
-        systemPrompt = systemPrompt.trim()
-        const answer = await complete(model, systemPrompt, prompt)
-        tui.print(answer)
-    })
-
-// TODO: generate with different aspect ratios
-aiCmd
-    .command('image')
-    .description('Generate a square image using an AI model')
-
-    .argument('<prompt>', 'description of the image to generate')
-    .option('-m, --model <model>', 'the graphic AI model to use', 'openai/dall-e-3')
-    .option('-s, --size <size>', 'the size of the image (in pixels)', '1024')
-    .option('-o, --output <path>', 'the output image path', 'image.png')
-
-    .action(async (prompt, { model, size, output }) => {
-        const sizeInt = z.int().min(16).max(2048).parse(convertStringToItsType(size))
-        const image = await generateImage(model, prompt)
-        await sharp(image).resize(sizeInt, sizeInt).toFile(output)
-        tui.success(`Generated image saved to ${output}`)
-        await tui.image(output, 20, 10)
-    })