diff --git a/doc/byoc-streaming.md b/doc/byoc-streaming.md
new file mode 100644
index 0000000000..2ef6076e7f
--- /dev/null
+++ b/doc/byoc-streaming.md
@@ -0,0 +1,171 @@
+# BYOC AI Stream API Documentation
+
+The AI Stream API provides real‑time video streaming with AI processing capabilities.
+All endpoints are rooted at **`/process/stream`** and are served by the Livepeer Gateway.
+
+**Base URL:** `https://{gateway-host}/process/stream`
+
+## Stream Setup
+
+_**Start**_ stream requests must include a `Livepeer` HTTP header containing a Base64‑encoded
+JSON object (the _Livepeer Header_). The header includes the job request, parameters,
+capability name and timeout. The client sends the header **unsigned**; the Gateway signs
+it before forwarding to Orchestrators.
+
+_**Update**_ requests also require a `Livepeer` header but reduced number of fields.
+See below.
+
+Routes that only require a valid stream_id in the URL path:
+
+- Ingest routes (WHIP and RTMP)
+- Status
+- Data
+
+Example `Livepeer` header JSON (before Base64 encoding):
+
+```json
+#Start Stream
+{
+  "request": "{}",
+  "parameters": "{\"enable_video_ingress\":true,\"enable_video_egress\":true, \"enable_data_output\":true}",
+  "capability": "video-analysis",
+  "timeout_seconds": 120
+}
+
+#Update Stream
+{
+  "request": "{\"stream_id\": \"254987ffsg\"}",
+  "parameters": "{}",
+  "timeout_seconds": 15
+}
+```
+
+The fields are defined as follows:
+
+- **timeout_seconds** — maximum processing time in seconds (integer, required).
+- **capability** — name of the AI capability to use (string, required).
+- **request** — JSON‑encoded `JobRequestDetails`.
+- **parameters** — JSON‑encoded `JobParameters`.
+
+### JobRequestDetails
+
+```json
+{
+  "stream_id": "string"
+}
+```
+
+- **stream_id** — ntifier of the stream to create or operate on (required).
+
+### JobParameters
+
+```json
+{
+  "orchestrators": {
+    "exclude": ["orch1_url", "orch2_url"],
+    "include": ["orch1_url", "orch2_url"]
+  },
+  "enable_video_ingress": true,
+  "enable_video_egress": true,
+  "enable_data_output": true
+}
+```
+
+- **orchestrators** — optional filter to select specific orchestrators.
+- **enable_video_ingress** — allow video input (boolean).
+- **enable_video_egress** — allow video output (boolean).
+- **enable_data_output** — enable server‑sent events data channel (boolean).
+
+## Data Types (JSON)
+
+### StartRequest
+
+```json
+{
+  "stream_name": "optional stream name that will be part of the stream_id returned",
+  "rtmp_output": "optional custom RTMP output URL",
+  "stream_id": "optional custom stream identifier",
+  "params": "JSON‑encoded string of pipeline parameters passed to worker"
+}
+```
+
+\*All StartRequest fields are optional; if `stream_id` is omitted a unique id will be generated.\_
+
+### StreamUrls
+
+```json
+{
+  "stream_id": "string",
+  "whip_url": "string (WebRTC WHIP ingest endpoint)",
+  "whep_url": "string (WebRTC WHEP egress endpoint)",
+  "rtmp_url": "string (RTMP ingest endpoint)",
+  "rtmp_output_url": "string (comma‑separated RTMP egress URLs)",
+  "update_url": "string (POST to update parameters)",
+  "status_url": "string (GET current status)",
+  "data_url": "string (SSE data channel, optional)"
+  "stop_url": "string"
+}
+```
+
+## Endpoints
+
+| Method   | Path                                | Description                                                                                  |
+| -------- | ----------------------------------- | -------------------------------------------------------------------------------------------- |
+| **POST** | `/process/stream/start`             | Create a new stream. Body = `StartRequest` JSON. Returns `StreamUrls` JSON.                  |
+| **POST** | `/process/stream/{streamId}/stop`   | Stop and clean up a running stream.                                                          |
+| **POST** | `/process/stream/{streamId}/whip`   | Endpoint to send ingest video via WebRTC WHIP (requires `LIVE_AI_WHIP_ADDR` set on Gateway). |
+| **POST** | `/process/stream/{streamId}/rtmp`   | Endpoint to send ingest video via RTMP.                                                      |
+| **POST** | `/process/stream/{streamId}/update` | Update stream parameters (`params` in request body passed to worker).                        |
+| **GET**  | `/process/stream/{streamId}/status` | Retrieve current stream status.                                                              |
+| **GET**  | `/process/stream/{streamId}/data`   | Server‑Sent Events endpoint for data channel output.                                         |
+
+### POST `/process/stream/start`
+
+- _Headers_: `Livepeer` (see Authentication).
+- _Body_: JSON `StartRequest`. All fields optional, but if `stream_id` is omitted the server generates one.
+- _Response_: JSON `StreamUrls` with all ingress/egress/update/status/data URLs and stream_id to use.
+
+### POST `/process/stream/{streamId}/update`
+
+- _Headers_: `Livepeer`. (requires `timeout_seconds` and `request` to be JSON encoded string of {"stream_id": streamId})
+- _Path_: `streamId` must reference an existing stream.
+- _Body_: JSON body passed to pipeline worker.
+- _Response_: HTTP 200 on successful parameter update.
+
+### POST `/process/stream/{streamId}/stop`
+
+- _Path_: `streamId` must reference an existing active stream.
+- _Body_: JSON body passed to pipeline worker.
+- _Response_: HTTP 204 No Content on success.
+
+### POST `/process/stream/{streamId}/whip`
+
+- _Path_: `streamId` must reference an existing stream.
+- _Environment_: `LIVE_AI_WHIP_ADDR` must be set.
+- _Response_: HTTP 200.
+
+### POST `/process/stream/{streamId}/rtmp`
+
+- NOTE: this is called by MediaMTX to signal an RTMP stream is being received. Client should use RTMP URL received in `/process/stream/start` response in streaming software.
+- _Path_: `streamId` must reference an existing stream.
+- _Response_: HTTP 200.
+
+### GET `/process/stream/{streamId}/status`
+
+- _Path_: `streamId` must reference an existing stream.
+- _Response_: JSON:
+
+```json
+{
+  "whep_url": "url",
+  "orchestrator": "https://current.orchestrator.url",
+  "ingest_metrics": {...metrics...}
+}
+```
+
+### GET `/process/stream/{streamId}/data`
+
+- _Path_: `streamId` must reference an existing stream.
+- _Response_: Server‑Sent Events stream delivering data‑channel messages (e.g., inference results) when `enable_data_output` is true.
+  Response is `text/event-stream` and streams JSONL from the pipeline worker container.
+  Response is `503` if no data output exists for stream