ros/claude-task-master
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(ai): Add xAI provider and Grok models

Browse Source 
Integrates the xAI provider into the unified AI service layer, allowing the use of Grok models (e.g., grok-3, grok-3-mini).

    Changes include:
    - Added  dependency.
    - Created  with implementations for generateText, streamText, and generateObject (stubbed).
    - Updated  to include the xAI provider in the function map.
    - Updated  to recognize the 'xai' provider and the  environment variable.
    - Updated  to include known Grok models and their capabilities (object generation marked as likely unsupported).
This commit is contained in:
Eyal Toledano
2025-04-27 14:47:50 -04:00
parent 2517bc112c
commit ed79d4f473
13 changed files with 315 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

89
.cursor/rules/ai_providers.mdc
View File

@@ -3,7 +3,6 @@ description: Guidelines for managing Task Master AI providers and models.
globs:
alwaysApply: false
---
# Task Master AI Provider Management
This rule guides AI assistants on how to view, configure, and interact with the different AI providers and models supported by Task Master. For internal implementation details of the service layer, see [`ai_services.mdc`](mdc:.cursor/rules/ai_services.mdc).
@@ -55,4 +54,90 @@ This rule guides AI assistants on how to view, configure, and interact with the
1. **Verify API Key:** Ensure the correct API key for the *selected provider* (check `models` output) exists in the appropriate location (`.cursor/mcp.json` env or `.env`).
2. **Check Model ID:** Ensure the model ID set for the role is valid (use `models` listAvailableModels/`task-master models`).
3. **Provider Status:** Check the status of the external AI provider's service.
4. **Restart MCP:** If changes were made to configuration or provider code, restart the MCP server.
4. **Restart MCP:** If changes were made to configuration or provider code, restart the MCP server.
## Adding a New AI Provider (Vercel AI SDK Method)
Follow these steps to integrate a new AI provider that has an official Vercel AI SDK adapter (`@ai-sdk/<provider>`):
1. **Install Dependency:**
- Install the provider-specific package:
```bash
npm install @ai-sdk/<provider-name>
```
2. **Create Provider Module:**
- Create a new file in `src/ai-providers/` named `<provider-name>.js`.
- Use existing modules (`openai.js`, `anthropic.js`, etc.) as a template.
- **Import:**
- Import the provider's `create<ProviderName>` function from `@ai-sdk/<provider-name>`.
- Import `generateText`, `streamText`, `generateObject` from the core `ai` package.
- Import the `log` utility from `../../scripts/modules/utils.js`.
- **Implement Core Functions:**
- `generate<ProviderName>Text(params)`:
- Accepts `params` (apiKey, modelId, messages, etc.).
- Instantiate the client: `const client = create<ProviderName>({ apiKey });`
- Call `generateText({ model: client(modelId), ... })`.
- Return `result.text`.
- Include basic validation and try/catch error handling.
- `stream<ProviderName>Text(params)`:
- Similar structure to `generateText`.
- Call `streamText({ model: client(modelId), ... })`.
- Return the full stream result object.
- Include basic validation and try/catch.
- `generate<ProviderName>Object(params)`:
- Similar structure.
- Call `generateObject({ model: client(modelId), schema, messages, ... })`.
- Return `result.object`.
- Include basic validation and try/catch.
- **Export Functions:** Export the three implemented functions (`generate<ProviderName>Text`, `stream<ProviderName>Text`, `generate<ProviderName>Object`).
3. **Integrate with Unified Service:**
- Open `scripts/modules/ai-services-unified.js`.
- **Import:** Add `import * as <providerName> from '../../src/ai-providers/<provider-name>.js';`
- **Map:** Add an entry to the `PROVIDER_FUNCTIONS` map:
```javascript
'<provider-name>': {
generateText: <providerName>.generate<ProviderName>Text,
streamText: <providerName>.stream<ProviderName>Text,
generateObject: <providerName>.generate<ProviderName>Object
},
```
4. **Update Configuration Management:**
- Open `scripts/modules/config-manager.js`.
- **`MODEL_MAP`:** Add the new `<provider-name>` key to the `MODEL_MAP` loaded from `supported-models.json` (or ensure the loading handles new providers dynamically if `supported-models.json` is updated first).
- **`VALID_PROVIDERS`:** Ensure the new `<provider-name>` is included in the `VALID_PROVIDERS` array (this should happen automatically if derived from `MODEL_MAP` keys).
- **API Key Handling:**
- Update the `keyMap` in `_resolveApiKey` and `isApiKeySet` with the correct environment variable name (e.g., `PROVIDER_API_KEY`).
- Update the `switch` statement in `getMcpApiKeyStatus` to check the corresponding key in `mcp.json` and its placeholder value.
- Add a case to the `switch` statement in `getMcpApiKeyStatus` for the new provider, including its placeholder string if applicable.
- **Ollama Exception:** If adding Ollama or another provider *not* requiring an API key, add a specific check at the beginning of `isApiKeySet` and `getMcpApiKeyStatus` to return `true` immediately for that provider.
5. **Update Supported Models List:**
- Edit `scripts/modules/supported-models.json`.
- Add a new key for the `<provider-name>`.
- Add an array of model objects under the provider key, each including:
- `id`: The specific model identifier (e.g., `claude-3-opus-20240229`).
- `name`: A user-friendly name (optional).
- `swe_score`, `cost_per_1m_tokens`: (Optional) Add performance/cost data if available.
- `allowed_roles`: An array of roles (`"main"`, `"research"`, `"fallback"`) the model is suitable for.
- `max_tokens`: (Optional but recommended) The maximum token limit for the model.
6. **Update Environment Examples:**
- Add the new `PROVIDER_API_KEY` to `.env.example`.
- Add the new `PROVIDER_API_KEY` with its placeholder (`YOUR_PROVIDER_API_KEY_HERE`) to the `env` section for `taskmaster-ai` in `.cursor/mcp.json.example` (if it exists) or update instructions.
7. **Add Unit Tests:**
- Create `tests/unit/ai-providers/<provider-name>.test.js`.
- Mock the `@ai-sdk/<provider-name>` module and the core `ai` module functions (`generateText`, `streamText`, `generateObject`).
- Write tests for each exported function (`generate<ProviderName>Text`, etc.) to verify:
- Correct client instantiation.
- Correct parameters passed to the mocked Vercel AI SDK functions.
- Correct handling of results.
- Error handling (missing API key, SDK errors).
8. **Documentation:**
- Update any relevant documentation (like `README.md` or other rules) mentioning supported providers or configuration.
*(Note: For providers **without** an official Vercel AI SDK adapter, the process would involve directly using the provider's own SDK or API within the `src/ai-providers/<provider-name>.js` module and manually constructing responses compatible with the unified service layer, which is significantly more complex.)*