AI Settings¶
AI settings control which AI provider is used for automatic content enrichment. The configuration is split into two sections that can be edited independently.
locali can automatically enrich imported content:
- Tagging: Automatically assigning tags based on content
- Area detection: Recognizing place names in text
- Summaries: Generating short content summaries (optional)
Section 1: Provider & Model¶
Click "Edit" next to Provider & Model to open the dialog.
| Field | Description |
|---|---|
| Provider | AI service: OpenAI, AzureOpenAI, GoogleAI, Cohere, Mistral, Ollama (self-hosted), … (config keys) |
| Model | The language model to use (e.g. gpt-4o-mini, llama3) |
| API endpoint | Base URL of the service (for Ollama: local address, e.g. http://localhost:11434) |
| API key | Authentication key — stored encrypted |
When an API key is stored, the overview shows "API key is saved" (the actual value is never displayed).
Section 2: Parameters¶
Click "Edit" next to Parameters to adjust model parameters.
| Field | Description |
|---|---|
| Temperature | Creativity / randomness (0 = deterministic, 1 = creative; default: 0.3) |
| Max. output tokens | Maximum length of the AI response (default: 4096) |
| Reasoning / Thinking | Enables extended reasoning for supported models |
| Ollama context length | Context window for Ollama models (optional, Ollama provider only) |
Provider overview¶
| Provider | Requirement | Best for |
|---|---|---|
| OpenAI | API key from openai.com | Simple cloud integration |
| Azure OpenAI | Azure resource + API key | Organisations with Azure environments and clear data-residency requirements |
| Google Gemini | API key from Google AI Studio | Flexible cloud use within Google ecosystem |
| Cohere | API key from cohere.com | NLP specialist models & embeddings |
| Mistral AI | API key from mistral.ai | European provider with EU data processing depending on setup |
| Ollama | Locally running Ollama instance | Data-sovereign operation without external AI service |
Cost awareness
AI enrichment generates API costs depending on provider and usage volume. For high-volume sources with frequent polling, review provider billing.
Review data protection and contracts
When external AI providers are used, content or questions may leave your own infrastructure. Operators must review provider choice, region, contracts, and legal basis themselves.
Fallback without AI
If no AI provider is configured, content is imported without AI tags. Keyword tag rules and routing rules work completely independently of AI.
Embedding model¶
For semantic search and "Ask locali", locali additionally requires an embedding model — a model that converts text into numerical vectors so that content similarity between questions and imported items can be recognised.
This is configured independently of the language model.
| Field | Description |
|---|---|
| Provider | Same provider as the language model, or a dedicated embedding service |
| Model | Recommended: text-embedding-3-small (OpenAI) or nomic-embed-text (Ollama) |
| Dimensions | Number of vector dimensions — depends on the model chosen |
Technical requirement
Embeddings are stored in the database. This requires the PostgreSQL extension pgvector to be active. It is already included in the standard Docker Compose configuration.
Activate "Ask locali"¶
"Ask locali" is the Q&A feature for members: they can ask questions in plain language and receive answers with source citations from the hub's imported content.
The feature relies on the language model and the embedding model — both must be configured and active.
| Setting | Description |
|---|---|
| Enable feature | Turns the Q&A feature on or off for members |
Answers should be grounded in the sources that were found and show those sources clearly. If no sufficient basis is available, the hub should not present a confident answer.
Requirement: embedding model
"Ask locali" only works when an embedding model is configured and all imported content has been indexed. Indexing runs automatically in the background — on a freshly set up hub this may take some time.