> ## Documentation Index
> Fetch the complete documentation index at: https://docs.banya.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Policy Configuration
> Learn how to register AI models, tool integrations, internal knowledge, development rules, and security policies.
Every settings page works the same way.
* Register items with the **Add** button; each item can be **edited or deleted**.
* Each page comes with **recommended presets** that you can apply instantly with a single click.
* For each item you set a **Required / Recommended** enforcement level. For details, see the [Policy operations guide](/en/admin/policies).
***
## AI Models
Register and approve the AI models your organization is allowed to use.
| Category | Description |
| -------------------- | ------------------------------------------------------------------------------------------ |
| **Built-in models** | Integrates with the local model server. Just toggle it on or off |
| **Supported models** | A catalog of models from major providers. Enable them individually or by group |
| **Custom models** | Models you register yourself (specify the endpoint, authentication method, and parameters) |
Supported integration types include official APIs, open-model APIs, cloud providers (Vertex AI · Azure OpenAI · AWS Bedrock), and self-hosted servers (OpenAI-compatible endpoints).
**Model whitelist** — Only the models approved by your organization appear in the model picker in the IDE. This lets you run a model strategy driven by **company cost and security standards** rather than individual preference.
***
## MCP Servers
Restrict integrations to only the external tools your organization has approved, such as GitHub and Jira.
* Register a server name and URL, and it is automatically deployed to every IDE in the organization.
* Authentication supports API keys or OAuth.
* Use **Test connection** and **Fetch tool list** to verify that everything works correctly.
Preserve extensibility while **preventing data leaks** caused by connecting unvetted external tools.
***
## RAG (Internal Knowledge)
Upload your internal wiki, API documentation, and policy documents, and the AI references them when answering.
| Action | Description |
| ------------------- | ---------------------------------------------------------------- |
| Create a RAG source | Upload documents along with a name and description |
| Add documents | Upload more documents to an existing source |
| Reindexing | Refresh the search index after documents change |
| Search settings | Adjust the similarity threshold and the number of search results |
Supported formats: `.txt` `.md` `.pdf` `.docx` `.json` `.csv`
By referencing shared team documents, the AI provides answers that **"understand your company's context."**
***
## Skills (Development Rules)
Register the team rules the AI must follow when writing code, in Markdown.
```
- Use camelCase for variable names
- Default JPA entities to FetchType.LAZY
- Wrap API responses in a Result wrapper object
```
Categories are divided into version control, coding style, project architecture, dependency policy, DB, and custom.
| Type | How it applies |
| --------- | -------------------------------- |
| **Rule** | Always included in the AI prompt |
| **Skill** | Loaded and used only when needed |
Developers don't have to memorize every rule; the AI **adheres to team standards** at the generation stage.
***
## Build/Test Settings
Register the validation and formatter commands to run automatically after code generation. You can specify different commands per language.
```
Validation: tsc --noEmit
Formatter: prettier --write .
```
If validation fails, the AI analyzes the error and attempts to fix it on its own.
***
## Hot Load
Register keywords, and when a team member types a sentence containing one of them in chat, a predefined command runs automatically.
| Item | Description |
| -------------------- | ------------------------------------------------------------------------------- |
| Trigger keywords | Register multiple keywords separated by commas |
| Command | The shell command to run |
| Completion condition | Exit code, whether the output contains a given string, or whether a file exists |
| Retries | Number of retries on failure |
| On failure | Stop, or pass the error to the AI to attempt a fix |
Turn your team's repetitive-task know-how into **automation rules as reusable assets**.
***
## Hooks
Attach shell commands to specific points in the workflow. Use them for audit logging or integrating with internal systems.
| Event | When it fires |
| --------------------- | --------------------------------------- |
| Before tool execution | Just before the AI runs a tool |
| After tool execution | After a tool finishes running |
| On tool failure | When a tool execution fails |
| Before LLM call | Just before a request is sent to the AI |
| On session start | When a new conversation begins |
You can set matching conditions so a hook responds only to specific tools, and configure synchronous/asynchronous execution and timeouts.
***
## Agents
Register subagents to deploy across the entire company. Team members invoke them with **`@agentname`** in the IDE chat.
| Item | Description |
| ------------- | ------------------------------------------- |
| Name | The invocation name that follows `@` |
| Description | What this agent does |
| System prompt | Its role and instructions |
| Allowed tools | The range of tools it can use |
| Read-only | Blocks file modification and deletion tools |
By deploying agents with dedicated roles—such as "security reviewer" or "test writer"—everyone on the team receives **reviews held to the same standard**.
***
## Security Rules
Block the execution of dangerous commands and access to sensitive files.
| Type | Description | Example |
| -------------------- | -------------------------------------------------- | --------------------------- |
| **Blocked commands** | Commands the AI cannot run | `rm -rf /` |
| **Protected files** | Files whose reading and modification is prevented | `.env*`, `credentials.json` |
| **Hidden files** | Files hidden so the AI cannot even know they exist | `config/production.json` |
***
## Exclude Patterns
Register the paths to exclude from project indexing, file search, and context collection as glob patterns. `.gitignore` is applied automatically.
```
.env*
node_modules/
dist/
```
Put **safeguards in place at the system level** without friction between the security and development teams.