> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zeptar.com/llms.txt
> Use this file to discover all available pages before exploring further.

# LLMs overview

> The large language models available to power an agent.

An **LLM** is the reasoning model an agent uses to understand input and decide what to say. The list is a curated set of models Zeptar runs through LiveKit Inference — there's no API to add or modify models.

The list is **flattened**: every model appears as one row, and each pinned dated snapshot appears as an additional row with `is_checkpoint: true`. A checkpoint carries the same provider and display name as its parent, but pins behavior to a fixed date so a deployment can't drift when the rolling alias updates.

Each entry reports capability hints — context window, max output tokens, image and document input support, parallel tool calls, and any reasoning-effort presets. `latency_hint_ms` and `cost_hint_per_min` are editorial display hints for the configuration UI, not authoritative latency or billing figures.

`default_deprecation_config` describes the lifecycle thresholds Zeptar applies when a model is retired and traffic is shifted to a fallback. Per-model `deprecation_info` is `null` until a specific model enters that lifecycle.

Use this endpoint to populate the model picker in the agent configuration UI.