feat: Add AI Agent workflow action node #12650
Conversation
…flowAiAgentActionSettings schema
…ith clearer labeling
…d AgentIdInput for GraphQL with validation
…ation and API key checks
…t retrieval and error handling
…improved agent management
…aceService, enhancing workflow step functionality
…eval, update, and deletion with feature flag checks
… for agent management with feature flag checks
…ntegrating TypeOrmModule and core modules for enhanced functionality
…ed functionalities
…ate and retrieval functionalities
…ate form state management and improve UI with CodeEditor for response format. Replace deprecated model provider and model selection logic with updated constants and hooks.
…odels, enhancing model selection options.
…ture flag checks in WorkflowStepResolver and AgentResolver
…ns hook for AI agent visibility
…tputSchema determination for HTTP requests
…roved localization support.
… handling, enhancing type safety and filtering based on API key availability.
…streamlining the test suite and focusing on valid scenarios.
… use factory functions for improved configuration and API key management.
…lId type and applying it to modelId property for AI models.
…ancing type safety in ClientConfig and related states.
… for improved localization support.
…nerated-metadata, ensuring consistency in module references.
…adability and maintainability.
...kflow/workflow-steps/workflow-actions/ai-agent-action/constants/output-field-type-options.ts
Outdated
Show resolved
Hide resolved
| name | ||
| description | ||
| prompt | ||
| modelId |
There was a problem hiding this comment.
Sorry should have made that feedback earlier sorry but modelId feels off, we usually use "name" in the codebase for slugs (identifier like "company") and label for end-user displayed name ("Company")
There was a problem hiding this comment.
Thanks for the feedback! I see the convention you're referring to, but I believe modelId is more accurate in this case. Values like 'gpt-4o' or 'claude-opus-4-20250514' are the actual model identifiers used by providers like openai and anthropic, not internal slugs like 'company'. Since this value is passed directly to the API to select the model, it's more of an external identifier than a conventional name.
There was a problem hiding this comment.
Ok I see that it's what AI SDK uses too so it makes sense 👍
packages/twenty-server/src/engine/core-modules/ai/utils/ai-model.utils.ts
Outdated
Show resolved
Hide resolved
packages/twenty-server/src/engine/core-modules/client-config/client-config.controller.spec.ts
Outdated
Show resolved
Hide resolved
packages/twenty-server/src/engine/core-modules/twenty-config/config-variables.ts
Outdated
Show resolved
Hide resolved
packages/twenty-server/src/engine/core-modules/twenty-config/config-variables.ts
Outdated
Show resolved
Hide resolved
| } | ||
|
|
||
| @Injectable() | ||
| export class FeatureFlagGuard implements CanActivate { |
…tions to streamline config variable definitions.
… across multiple files for consistency and clarity in the ClientAIModelConfig structure.
…ent by removing unnecessary translation macro, enhancing code clarity.
…el' for AI models, ensuring consistency with recent configuration changes.
…elector for improved localization support.
|
Hi, We upgraded to latest v1.03 but still don't see the AI Agent in workflow. Please advise when it will be released? |
|
Hey @salesbay it's behind a feature flag IS_AI_ENABLED which you can set in the admin panel if you're self-hosting. We still need to iterate on it, it's going to be a lot more powerful :) |
Thanks, Felix! We enabled it in the .env docker file and in the docker-compose.yml, but we can't see it in the Admin panel. |
agent.mov