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

# Glossary

> Canonical definitions for Overshoot terms used across the docs and API.

Use these terms consistently when reading or writing about Overshoot. The source of truth for the wire format is [`api-reference/openapi.yaml`](/api-reference/introduction).

## Stream

A leased live-video session. A Stream holds the most recent window of frames your client publishes (up to roughly 10 minutes of buffer) and exposes them to vision-language models through the `ovs://` URI scheme.

* Created by [`POST /streams`](/api-reference/create-stream).
* Inspected by [`GET /streams/{stream_id}`](/api-reference/get-stream).
* Renewed by [`POST /streams/{stream_id}/keepalive`](/api-reference/keepalive-stream).
* Deleted by [`DELETE /streams/{stream_id}`](/api-reference/delete-stream).

A Stream ends when the lease expires (5 minutes idle) or the client deletes it. See [The Stream](/the-stream) for the full lifecycle.

## `stream_id`

UUID returned by `POST /streams`. Use this exact spelling in prose and code — not `"streamId"` or `"stream ID".` Pass it as a path parameter on stream endpoints and as part of the `ovs://` URI.

## Frame

A single image extracted from the publisher's video track and stored in the Stream's buffer. Reference an individual frame from chat completions with `frame_index` (see below).

## `frame_index`

Integer index into a Stream's retained frames.

* `frame_index=-1` — the most recent frame (typical for "what's happening right now").
* `frame_index=0` — the first frame still retained in the buffer (older frames are evicted).

Returned values include `last_frame_index`, `first_available_frame_index`, `retained_frame_count`, and `evicted_frame_count` from [`GET /streams/{stream_id}`](/api-reference/get-stream).

## Segment

A time-bounded slice of a Stream, referenced via `start_offset_ms` and optionally `end_offset_ms`.

* `start_offset_ms=-5000` means "5 seconds ago relative to the live edge".
* `start_offset_ms=0` means "the start of the available buffer".
* Omit `end_offset_ms` to mean "now".

## `ovs://` URI

Overshoot's stream-reference URI scheme. Used inside `image_url` or `video_url` content parts on `/chat/completions`.

* Frame: `ovs://streams/{stream_id}?frame_index=-1`
* Segment: `ovs://streams/{stream_id}?start_offset_ms=-5000`
* Bounded segment: `ovs://streams/{stream_id}?start_offset_ms=0&end_offset_ms=10000`

The `ovs://` URI is a reference identifier the server parses to extract `stream_id` and the query — it is not a fetchable URL. See [Chat Completion](/chat-completion) for examples.

## Lease

The keepalive window for a Stream. Default is 5 minutes from the most recent keepalive (or stream creation). When the lease expires, the Stream is deleted.

## Keepalive

The act of renewing a Stream's lease. Call [`POST /streams/{stream_id}/keepalive`](/api-reference/keepalive-stream) roughly every 2 minutes while the Stream is in use. Each keepalive returns a fresh LiveKit token you can use to rejoin the room if your publisher disconnects.

## Model id

The string you pass as the `model` field on `/chat/completions`. Format is `provider/model-name`:

* `Qwen/Qwen3.6-27B-FP8`
* `Qwen/Qwen3.6-35B-A3B-FP8`
* `google/gemma-4-26B-A4B-it`
* `google/gemma-4-31B-it`
* `Hcompany/Holo3-35B-A3B`

List currently-ready model ids with [`GET /models`](/api-reference/list-models) or browse the [Models page](/models).

## LiveKit room

The WebRTC publish target returned by `POST /streams`. Comprises a `wss://` URL and a short-lived JWT token. Connect with any [LiveKit SDK](https://docs.livekit.io/reference/) to publish your video track. Overshoot's media gateway is the only other participant in the room.

## API key

Bearer token used to authenticate requests. Prefixed with `ovs-`. Issued from the [Overshoot dashboard](https://overshoot.ai/dashboard). Only [`GET /models`](/api-reference/list-models) is unauthenticated.