# Models & Playground

Seclai exposes AI models in two different ways:

1. A **public models catalog** for browsing supported LLMs, embedding models, and rerankers.
2. An **account-level models page** where authenticated users can launch an LLM playground and test models directly.

This page explains how those two experiences fit together, plus what is currently available through the REST API and MCP server.

## UI Flow

### Public catalog

Visit `/models` to browse the full model catalog without signing in.

- **LLMs** show provider, context window, output limits, and capability badges. Models with the **tool-use** badge support function calling and can use all of Seclai's platform-provided tools — including web search, web fetch, knowledge base retrieval, content inspection, and memory banks — regardless of whether the vendor offers those capabilities natively.
- **Embedding models** show dimensions, credits, and language support.
- **Rerankers** show provider, credits, and default status.

For LLMs, each model detail page provides a **Try This** action.

- If you are signed in, Try This forwards you to the Model Experiments page with the selected model prefilled.
- If you are signed out, Seclai redirects you to login first and then returns you to the account resolver flow.

### Account models page

Visit `/app/{account_id}/models` after signing in.

This page is the authenticated version of the model catalog. It keeps the catalog visible but adds model experiment actions:

- **Model Experiments** is a dedicated page for testing and comparing models side by side.
- Starting from public model details via Try This pre-fills the selected model and prompt.

## Playground Editor Modes

### Simple Editor

Use the **Simple** editor when you want the familiar prompt-call flow:

- A user prompt template
- An optional system template
- Manual substitution inputs
- A quick response preview from the selected model

This is the fastest way to try a model for normal conversational or instruction-following prompts.

### Advanced JSON Editor

Use the **Advanced** JSON editor when you need more control over the payload.

This mode is useful when:

- You want to send structured message arrays
- You need model-specific JSON fields
- You want to validate a more advanced prompt-call configuration before using it in an agent

## Evaluation Depth

When using **AI Evaluator** in the playground, Evaluation Depth controls how much analysis the evaluator performs.

| Depth        | What it does                                   | Best for                              |
| ------------ | ---------------------------------------------- | ------------------------------------- |
| **Simple**   | Fast pass/fail style scoring and quick ranking | Rapid comparisons and low-cost checks |
| **Standard** | Multi-criteria scoring with clearer tradeoffs  | Everyday side-by-side model testing   |
| **Complex**  | Most rigorous analysis with deeper reasoning   | High-stakes prompt quality reviews    |

In general, start with **Standard**, move to **Simple** for quick iterations, and use **Complex** when precision matters more than speed.

## API Access

There are two separate API concepts related to models:

### Model metadata APIs

Seclai already exposes REST endpoints for model metadata and lifecycle information, such as model recommendations and model lifecycle alerts. See [Model Lifecycle](https://seclai.com/docs/model-lifecycle) for those endpoints.

### Prompt-call testing

The UI playground uses an authenticated prompt-call test flow under the hood. This is useful for testing a model with prompt templates and substitutions before saving the configuration into an agent.

For API-first workflows, the practical guidance is:

- Use the normal REST API or SDKs to create/update agent prompt-call steps for durable configurations.
- Use the authenticated prompt-call testing flow when you specifically want ad hoc evaluation behavior like the UI playground.

If you are integrating programmatically and need stable automation, the agent-definition APIs are the better long-term entry point than treating the UI playground as your primary workflow.

## MCP Access

The MCP server currently exposes model **lifecycle** tooling, not a standalone model playground tool.

Today, MCP supports model-related workflows such as:

- Listing model lifecycle alerts
- Fetching replacement recommendations for deprecated or sunset models

See [MCP Server](https://seclai.com/docs/mcp) and [Model Lifecycle](https://seclai.com/docs/model-lifecycle) for the current tool set.

If you are using MCP and want to test prompt behavior, the recommended approach today is still to work through agent configuration or the UI playground rather than a dedicated MCP playground command.

## Recommended Workflows

### I just want to explore what models exist

Browse the public [Models](https://seclai.com/models) catalog — no sign-in required.

### I want to try prompts against a model before editing an agent

Open **Model Experiments** in your account sidebar, or click **Try This** on any model in the public catalog to jump straight to the experiment page with that model prefilled.

### I want to make a durable prompt-call configuration

Use the [agent editor](https://seclai.com/docs/agents) in the UI, the agent-definition REST API, the [MCP server](https://seclai.com/docs/mcp) tools, or an SDK.

### I want automated model lifecycle management

Use the UI alerts in **Models → Lifecycle**, the REST API, or [MCP](https://seclai.com/docs/mcp) tools documented in [Model Lifecycle](https://seclai.com/docs/model-lifecycle).

## Next Steps

- [Embedding Models](https://seclai.com/docs/embedding-models) — Learn about embedding and reranker model choices
- [Model Lifecycle](https://seclai.com/docs/model-lifecycle) — Recommendations, deprecations, and lifecycle alerts
- [API Introduction](https://seclai.com/docs/api-introduction) — Authenticate and integrate through REST
- [MCP Server](https://seclai.com/docs/mcp) — Use Seclai from MCP-compatible tools