> ## Documentation Index > Fetch the complete documentation index at: https://docs.open-metadata.org/llms.txt > Use this file to discover all available pages before exploring further. # AI Context > Choose which metadata is compiled into an AI context document for each persona, so Model Context Protocol (MCP) assistants answer from the data that matters most to each role. # AI Context **AI Context** lets administrators define rules for each persona. The rules select which metadata is compiled into a context document for that persona. When a user with that persona connects an MCP assistant, the assistant can read the compiled document with the [`get_persona_context`](https://docs.open-metadata.org/v2.1.x-SNAPSHOT/how-to-guides/mcp/reference#get_persona_context) tool. Answers are then based on the metadata that matters most to the user's role, not only on what search surfaces for a single question. Some rules can also narrow the search results an assistant sees. For more information, see [Filtered in Search](#filtered-in-search). For example: * A **business analyst** persona gets certified, top-tier tables, so the assistant starts from trusted data. * A **data engineer** persona gets broad coverage across all services, so nothing is left out. * A **compliance** persona always gets regulatory Glossary Terms and policy Articles, so the assistant has the right definitions and constraints. > [!NOTE] > > Only administrators can manage AI Context. ## Open AI Context 1. From the left navigation bar, go to **Context Center** > **AI Context**. The page lists every persona. 2. Select a persona to open its AI Context settings and rules. The persona list has one row per persona with these columns: * **Persona**: The persona name. * **Users**: How many users have the persona. * **Rules**: How many rules the persona has. * **Entities**: How many entities the compiled context includes. * **Context budget used**: How much of the persona's budget the compiled context uses, shown as estimated tokens out of the total. It shows **No context generated** until the context is compiled. * **Freshness**: The state of the compiled context. For more information, see [Freshness](#freshness). * **Enabled**: A toggle that turns AI Context on or off for the persona without opening it. ![AI Context persona list with Persona, Users, Rules, Entities, Context budget used, Freshness, and Enabled columns](https://mintcdn.com/openmetadata/APiubSIYs8BN3RC7/public/images/how-to-guides/context-center/ai-context-persona-list.png?fit=max\&auto=format\&n=APiubSIYs8BN3RC7\&q=85\&s=f145095a9f9fcb38eea6316b30ed49c5) ## Global Settings At the top of the persona's AI Context page, set the overall behavior before you add rules. | Setting | Description | | - | - | | **Enabled** | Turns context compilation on or off for the persona. When it is off, the persona's rules and prompt have no effect. | | **Character budget** | The maximum number of characters in the compiled context document. Default: 400,000. Allowed range: 10,000 to 2,000,000. | | **Cache TTL** | The time to live (TTL): how long, in minutes, the compiled context stays fresh before it is compiled again. Default: 30. Allowed range: 1 to 1,440. A lower value keeps the context current. A higher value means fewer recompilations. | ![AI Context global settings: Enabled toggle, Character budget, Cache TTL, and freshness status](https://mintcdn.com/openmetadata/APiubSIYs8BN3RC7/public/images/how-to-guides/context-center/ai-context-global-settings.png?fit=max\&auto=format\&n=APiubSIYs8BN3RC7\&q=85\&s=7e008d827bedf9440cbc0766d6838705) Rules are compiled in order, and rules marked **Always in context** come first. An entity that matches more than one rule appears only under the first rule that matches it. > [!NOTE] > > The character budget is a target, not a hard cap. Content that is always included, such as Articles from **Always in context** rules, can exceed it. ### Version History The version indicator, for example **v0.6**, shows the current version of the persona's AI Context. Select it to open the version history. The history lists every change. Examples include rules added, updated, or deleted; the prompt updated; the budget or Cache TTL changed; and AI Context enabled or disabled. To go back to an earlier version, select **Restore** next to it and confirm. Restoring creates a new version with the rules and settings of the version you chose. ### Freshness Next to the settings, the page shows when the context was last compiled, for example **Last generated 5 minutes ago**, with a status badge. If the context was never compiled, it shows **Context has not been generated yet**. If compilation fails, the page shows the reason. ![AI Context freshness status next to the global settings](https://mintcdn.com/openmetadata/APiubSIYs8BN3RC7/public/images/how-to-guides/context-center/ai-context-freshness.png?fit=max\&auto=format\&n=APiubSIYs8BN3RC7\&q=85\&s=2eb4e36a81133c22a42f33a0c8daa578) On the persona list, **Freshness** shows one of these states: * **Fresh**: The compiled context is current. The badge also shows when it was generated. * **Stale**: The compiled context is out of date and is compiled again when it is next needed. * **Not generated**: The persona has rules, but the context hasn't been compiled yet. * **Generating**: The context is being compiled. * **Failed**: The context couldn't be compiled. * **Not configured**: The persona has no rules yet. ## Add a Prompt The **Prompt** box is optional. Use it to tell the assistant who the users of this persona are and what they are responsible for. For example: *You are assisting business users at Acme. They understand revenue and churn metrics, and they are responsible for quarterly planning.* ![AI Context Prompt box with Cancel and Save buttons](https://mintcdn.com/openmetadata/APiubSIYs8BN3RC7/public/images/how-to-guides/context-center/ai-context-prompt.png?fit=max\&auto=format\&n=APiubSIYs8BN3RC7\&q=85\&s=1d48ac413efe9a6d537f68fa914e2093) 1. Enter the instructions in **Prompt**. The prompt can be up to 8,000 characters. 2. Click **Save**. Click **Cancel** to discard your changes. The prompt is returned with the first part of the persona's context as instructions, while AI Context is enabled. It helps the assistant tailor answers to the persona's users. It doesn't override the assistant's own rules. Saving the prompt is recorded in the version history. ## Add a Rule Each rule selects a set of entities whose metadata is compiled into the context document. Add several rules to build the full context for a persona. For example, use one rule for core tables, another for Glossary Terms, and another for policy Articles. 1. Click **Add Rule**. ![Add Rule button on the AI Context page](https://mintcdn.com/openmetadata/APiubSIYs8BN3RC7/public/images/how-to-guides/context-center/ai-context-add-rule.png?fit=max\&auto=format\&n=APiubSIYs8BN3RC7\&q=85\&s=03e7962c2d3de9cf8c7b07f94a9924a6) 2. Fill in the rule fields. For more information, see [Rule Fields](#rule-fields). ![Add Rule panel with Name, Description, Entity Type, and Filter](https://mintcdn.com/openmetadata/APiubSIYs8BN3RC7/public/images/how-to-guides/context-center/ai-context-configure-rule.png?fit=max\&auto=format\&n=APiubSIYs8BN3RC7\&q=85\&s=6b9bfabd81c6e15b44a71210322b43c1) 3. Click **Save Rule**. When you add a rule for a data asset, **Filtered in search** starts on. When you add a rule for a Glossary Term, Article, or Metric, **Always in context** starts on. You can change either before you save. To change a rule, click its edit icon. To remove it, click its delete icon and confirm. ## Rule Fields Each rule has the following fields. ### Name and Description Give the rule a name that says what it selects, for example *Core finance tables* or *Compliance policy Articles*. The name must be unique within the persona. The description is optional. ### Entity Type Choose the type of entity the rule selects. Each rule selects one type, so create separate rules to include different types. * **Data assets**: Table, Topic, Dashboard, Chart, Data Model, Pipeline, ML Model, Container, Database, Database Schema, Stored Procedure, Search Index, API Collection, API Endpoint, and Data Product. * **Knowledge**: Glossary Term, Article, and Metric. ### Filter Add conditions to narrow which entities the rule includes. Build them the same way as in the Explore advanced filter, using any field available for the entity type. Without a filter, the rule selects every entity of the chosen type. As you build the filter, the rule shows how many entities match, with sample names. Click **View in Explore** to check the selection before you save. ![Rule filter with AND and OR conditions and the matched entities count](https://mintcdn.com/openmetadata/APiubSIYs8BN3RC7/public/images/how-to-guides/context-center/ai-context-rule-filter.png?fit=max\&auto=format\&n=APiubSIYs8BN3RC7\&q=85\&s=a49a0b0e371d8f248c1d34ceaebf361c) > [!NOTE] > > An unfinished condition is ignored when the rule runs, which would widen the rule to match everything. The editor asks you to finish or remove it before you can save. ### Always in Context When on, the rule's content is added to every request for the persona, whatever the user asks about. This works well for Articles. For example, turn it on for a rule that selects Articles with domain conventions, business definitions, or team standards, so the assistant always has them. ### Filtered in Search When on, the entities the rule selects are not preloaded into the context document. Instead, the rule becomes a default filter on the assistant's metadata searches for the persona, so the persona sees only the assets the rule selects. Because nothing is preloaded, the rule doesn't use the character budget. When this is on, **Always in context**, **Fully rendered**, **Context sections**, and **Max assets** don't apply. Rules for Glossary Terms, Articles and Metrics are always preloaded with their full content. **Filtered in search** isn't available for them. ### Fully Rendered When on, each matched entity's full documentation is compiled into the context. For a Data Product, this also includes the context of every asset it contains. When off, only the sections you pick under **Context sections** are included. Turn this on only when the persona needs complete documentation. It uses much more of the character budget. Rules for Glossary Terms, Articles and Metrics are always fully rendered. ### Context Sections When **Fully rendered** is off, choose which parts of each entity's documentation are compiled. The sections depend on the entity type. For data assets, the sections are: | Section | Default | Notes | | - | - | - | | **Description** | On | | | **Schema** | On | | | **Keys and Constraints** | On | | | **Join** | On | | | **Tags** | On | | | **Related Glossary Terms** | On | | | **Related Articles** | On | | | **Related Metrics** | On | | | **Lineage** | Off | Marked **heavy**. Adds a lot of content. | | **Profile** | Off | Marked **heavy**. Adds a lot of content. | | **Data Quality** | Off | Marked **heavy**. Adds a lot of content. | Turn on the heavy sections only when the persona needs that depth. They use a large share of the character budget and can use budget that other rules need. For other entity types, the sections are: | Entity type | Sections | On by default | | - | - | - | | **Glossary Term** | Definition, Synonyms, Related Terms, Tags, Related Assets | Definition | | **Article** | Title and Summary, Full Body, Tags, Related Glossary Terms, Related Assets | Title and Summary, Full Body, Tags | | **Metric** | Definition, Formula Expression, Unit and Grain, Owner, Tags, Related Assets | Definition, Formula Expression, Unit and Grain | ### Max Assets The maximum number of entities that the rule compiles. Default: 200. Allowed range: 1 to 1,000. When more entities match, the highest-priority ones are compiled first. Lower this value to leave more budget for other rules. ## Preview the Compiled Context Click **Preview Context** to see the context document as an assistant receives it. The preview shows how many entities are included, how many were truncated, the estimated size in tokens and KB, and when it was generated. Use **Rendered** and **Raw** to switch views, **Refresh** to compile it again, and **Copy** to copy the content. The **Content** list jumps to each rule and the entities it includes. ![Preview Context window showing the compiled context for a persona, with the Content list, Rendered and Raw views, Refresh, and Copy](https://mintcdn.com/openmetadata/APiubSIYs8BN3RC7/public/images/how-to-guides/context-center/ai-context-preview.png?fit=max\&auto=format\&n=APiubSIYs8BN3RC7\&q=85\&s=200408d5c4f0b0686c318d739712bc2b) Use the preview to confirm the right tables, terms, Articles, and metrics are included and that the budget is used well. ## Example Configurations **Business analyst persona** * Rule 1: Tables, filtered by Tier 1 and a certification, with **Filtered in search** off. This includes only trusted, business-critical assets in the context. * Rule 2: Glossary Terms, filtered by the analyst's domain. The assistant uses agreed business definitions. * Rule 3: Articles, filtered by a tag such as `analyst-conventions`, with **Always in context** on. Domain conventions are always available. **Data engineering persona** * Rule 1: Tables, no filter, with **Filtered in search** off and a high **Max assets** value. This gives broad coverage across services. * Rule 2: Articles, filtered by a tag such as `pipeline-runbooks`, with **Always in context** on. Runbooks are always included. **Compliance persona** * Rule 1: Tables, filtered by a PII tag, with **Filtered in search** off. This includes sensitive data assets first. * Rule 2: Glossary Terms, filtered by a regulatory classification. Correct definitions are always available. * Rule 3: Articles, filtered by a tag such as `compliance-policy`, with **Always in context** on. Policy documents are included in every interaction. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.