> ## Documentation Index > Fetch the complete documentation index at: https://docs.parable.work/llms.txt > Use this file to discover all available pages before exploring further. # Workspace AI processors > See which AI processors receive your Workspace's data and on what terms, switch processors off after seeing what stops, choose the models your features use, and order the models Ponder offers. # Workspace AI processors ## Sources Docs: [Security and privacy](/protection/security-privacy.md), [Ponder Primer](/ponder/primer.md) (the model menu, what Ponder shows when chat is unavailable, Clean up text and Keep the Parable named), [Protection Primer](/protection/primer.md). Web App: Workspace administrators open **Protection → AI Processors**. Some Parable features run an AI model over your Workspace's data: Ponder chat, Ponder's Clean up text, Parable naming and judgment tools, the AI functions in SQL, semantic Pool search, and schema inference on uploads. An AI processor is the service that runs the model. **Protection → AI Processors** lists every processor that can receive your Workspace's data, with its retention terms, and lets a Workspace administrator switch processors off, choose the models features use, and order the models Ponder offers. ## Who can change it | Who | What they change | | - | - | | Workspace administrators | Switch processors off and back on, set the Workspace's default models, and order and hide **Ponder models**. | | Parable staff | Which processors your Workspace may use at all, and a Parable-owned Google Cloud project for your Vertex AI calls. | Other members of the Workspace cannot open the page. A processor that Parable has not allowed for your Workspace shows **Not allowed by Parable**, and you cannot switch it on. ## The processors | Processor | What Parable uses it for | Retention | | - | - | - | | Google Vertex AI | Ponder chat on Ponder and on the Claude and Gemini models Parable hosts on Vertex AI, `embed()`, `ponder()` and `model()` in SQL, semantic Pool search, search of Provider records by meaning, and schema inference on uploads | Google Cloud's service terms for Vertex AI, under Parable's Google Cloud agreement. Calls run in a Parable-owned Google Cloud project. | | OpenRouter | Ponder chat on a model chosen from OpenRouter's list, and Ponder naming a Parable from its Messages (**Keep the Parable named**) | Every request requires zero data retention and denies provider data collection, so only model endpoints that keep no prompts or outputs serve it. | | Anthropic | No feature sends it your Workspace's data. | Anthropic's commercial API terms. No zero-data-retention agreement with Anthropic is on record. | | TypeSafe | Jev in SQL, the Ponder composer's **Clean up text** (on unless a person turns it off), and Ponder's judgment tools | TypeSafe's service terms. No zero-data-retention agreement with TypeSafe is on record. | The page is the current record of the processors and terms for your Workspace. The same processors appear on the sub-processor list in the [Trust Center](https://trust.askparable.com/). ## Switch a processor off 1. Open **Protection → AI Processors**. 2. Turn off the processor's switch. 3. Read **Before you save**. Parable works out what the change would do and lists each feature that stops. For example, switching Vertex AI off stops Ponder chat on Ponder and on every model Parable hosts on Vertex AI, makes Pool search match keywords only, refuses `embed()` in SQL, and stores uploads without AI-inferred column metadata. Switching OpenRouter off stops chat on the models OpenRouter serves. 4. Choose **Save changes** in the page header, or **Discard**. Once a processor is off, Parable sends it none of your Workspace's data for the features that follow these settings. A feature that needs a switched-off processor stops with a clear message; it never falls back to another processor. Ponder chat, Ponder's language features, the SQL AI functions, Pool search and schema inference on uploads all follow these settings. Schema inference on uploads runs on Google Vertex AI. With Vertex AI off, uploads still succeed; they are stored without AI-inferred column metadata. ## Choose default models **Default models** has one row per feature you can set. A default is the model a feature uses when nothing names one: a call that names its own model, such as `ponder()` with explicit model arguments, always uses that model. 1. Turn on **Use a Workspace default** for the feature. 2. Choose the model: * `ponder()` and `model()` take a processor and the model's provider, name, variant and version. * Jev takes a Jev model version. * Schema inference takes a processor and its model id. 3. Read **Before you save** and save. Turn a feature's switch off to go back to Parable's default. Changing the default for `ponder()`, `model()` or Jev changes what scheduled Plots compute. Before you save, the review counts the scheduled Plots that will recompute on their next run: the Plots your Report schedules run whose SQL calls the function without naming a model. For Jev it also counts the Plots that may recompute, because they take the model from a column and rows that leave it empty use the default. When Parable cannot count them, the review says that scheduled Plots calling the function without naming a model recompute on their next run. Ponder chat has no Workspace default model. The composer starts on Ponder Standard, and each person chooses a model there. ## Ponder models **Ponder models** lists the models Ponder's composer offers, in the order it offers them. Each row shows the model, the server that answers for it, and whether it is available in this Workspace with the reason when it is not. Models Parable hosts show **Parable** as their server, whichever processor runs them; the processors table above names the processor. 1. Open **Protection → AI Processors** and find **Ponder models**. 2. Drag a row by its handle to change the order, or pick the handle up with Space and move it with the arrow keys. 3. Hide a model to take it out of the composer's model menu, or show it again. 4. Choose **Save changes** in the page header, or **Discard**. The order and hidden models apply to everyone in the Workspace. Hiding a model does not switch off its processor, and showing a model does not make it available: the Workspace's processors still decide which models can answer. ## The embedding model Parable sets the embedding model for every Workspace, so vectors stored for your data stay comparable. The page shows it under **Embedding model** for reference; it cannot be changed from the Workspace. **Search of Provider records by meaning** embeds the text of your Provider records, separately from Pool search, which embeds table and column descriptions. Switching a processor off can stop one without the other. This search runs on Gemini embedding on Google Vertex AI, so switching Google Vertex AI off stops it. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.