Configuration
The Parsium Settings page has four tabs. Open it by clicking the Parsium Settings tab in your Salesforce navigation. Here is what each tab does and how to configure it.
Credentials Tab
Tab 1This is where you select and test your Named Credential connection.
- Named CredentialA dropdown showing all Named Credentials in your org. Select the one you created during setup (default:
Parsium_OpenRouter). You can search by typing in the dropdown. - Test ConnectionClick this button after selecting a credential. Parsium will make a test call to OpenRouter. A green "Connected" badge means everything is working. A red badge means something is wrong (check the error message).
- Credential DetailsAfter selecting a credential, Parsium shows the endpoint URL and associated External Credential name so you can verify the configuration.
Privacy & Security Tab
Tab 2Controls how your data is handled by AI providers. See Privacy Levels below for full details on each level.
- Privacy LevelA three-way toggle: Standard, Enhanced, or Maximum. Higher levels restrict which AI providers can handle your data. Default is Maximum.
- Request LoggingA checkbox that controls whether the full prompt content (your instructions + extracted text) is saved in Usage Log records. Useful for debugging, but disable it in production if your documents contain sensitive data.
When Maximum privacy is selected, the model list in the Model Selection tab will only show models available through Zero Data Retention (ZDR) endpoints. Some models may not be available at this level.
Model Selection Tab
Tab 3Browse and select the AI model that will process your documents. See AI Models & Parameters below for guidance on choosing the right model.
- Model BrowserA sortable table showing all available models. Columns include Company, Name, Context Length, and Input/Output pricing. Click any column header to sort.
- Capability FiltersFilter buttons at the top let you narrow models by what they support: Files, Images, Audio, or Video. Only models with the selected capability are shown.
- SearchType to filter models by name or provider. Results update as you type.
- TemperatureThree presets: Precise (0), Balanced (0.5), Creative (1). Lower values make the AI more deterministic. Default is Precise (0).
- Reasoning EffortOnly visible for models that support reasoning. Three levels: Low, Medium, High. See Reasoning Effort.
Field Mapping Tab
Tab 4Define which Salesforce fields should be populated by the AI and what instructions to give it. See the Field Mapping Guide for a full walkthrough.
- Object SelectorA searchable dropdown listing every Salesforce object you have permission to update. Select the object you want to create mappings for.
- Field TableAfter selecting an object, a table appears showing your current field mappings. Each row has a Field, Type badge, Instruction, and Delete button.
- Add FieldClick "Add Field" to add a new mapping row. Select a field from the dropdown, then write an extraction instruction. Already-mapped fields are excluded from the dropdown.
- Save AllSaves all field mappings for the selected object in a single batch. You must click Save after making changes.
Privacy Levels
Parsium offers three privacy levels that control how AI providers handle your data. The default is Maximum, the most restrictive option.
Standard
Level 0Most models availableNo restrictions on data handling. AI providers may use your data according to their standard terms, which could include training future models. This gives you access to the widest selection of AI models at the lowest cost.
When to use: Use this for non-sensitive data where cost and model selection are more important than data privacy. Good for testing and development.
Enhanced
Level 1Good model selectionParsium sends a data_collection: "deny" flag with every request, telling the AI provider not to use your data for training. Most major providers honor this flag. You still get a good selection of models.
When to use: Use this for business data that is not highly regulated but should not be used to train AI models. A good middle ground for most organizations.
Maximum
Level 2ZDR endpoints onlyEverything from Enhanced, plus Parsium only shows models available through Zero Data Retention (ZDR) endpoints. ZDR means the provider guarantees your data is never stored, logged, or retained after the response is generated.
When to use: Use this for regulated industries (healthcare, finance, legal), sensitive documents, or any time data retention by a third party is unacceptable. This is the default setting.
What happens to your data when it's sent to an AI model?
Peer-reviewed research from Google DeepMind, Meta, and leading AI security researchers shows that your data is statistically impossible to memorize or retrieve from a trained model. We wrote a detailed breakdown with 14 cited sources.
Read: Is Your Data Safe with AI?AI Models & Parameters
Parsium connects to AI models through OpenRouter, which gives you access to models from OpenAI, Anthropic, Google, Meta, DeepSeek, and many others.
Choosing a Model
| Factor | What It Means |
|---|---|
| File support | Not every model can process every file type. Use the capability filters (Files, Images, Audio, Video) to find models that support your file types. |
| Context length | How much text the model can handle in a single request. Large documents need models with 100k+ tokens. |
| Pricing | Models charge per token (~4 characters = 1 token). Smaller models like GPT-4o Mini are much cheaper than GPT-4o or Claude Opus. |
| Quality | Larger models produce more accurate extractions for complex documents. For simple invoices, a small model works fine. |
| Speed | Smaller models respond faster. If processing time matters, choose a smaller model. |
Recommended starting point: OpenAI GPT-4o is the default. It supports images, has a large context window, and offers a good balance of quality, speed, and cost. Switch to GPT-4o Mini for faster, cheaper extractions on simple documents.
Temperature
Temperature controls how "creative" the AI model is. For data extraction, you almost always want a low temperature.
| Preset | Value | Best For |
|---|---|---|
| Precise | 0 | Data extraction, invoices, forms, structured data (recommended) |
| Balanced | 0.5 | General use, mixed content documents |
| Creative | 1 | Summarization, descriptions, free-text fields |
Reasoning Effort
Some AI models support "reasoning", which means they think step-by-step before giving an answer. This improves accuracy for complex tasks but takes longer and costs more.
| Level | Value | Description |
|---|---|---|
| Low | 1 | Minimal reasoning. Fastest response, lowest cost. |
| Medium | 2 | Moderate reasoning. Good balance of accuracy and speed. |
| High | 3 | Maximum reasoning. Slowest but most thorough for complex documents. |
Models That Support Reasoning
New reasoning models are released frequently. Parsium automatically detects which models support reasoning and shows the effort toggle only when applicable. You can browse the current list of reasoning-capable models on OpenRouter:
View reasoning models on OpenRouter