Trio Stream Vision — Analyze Any Livestream with Natural Language

Paste a YouTube Live URL, RTSP camera feed, or HLS stream — ask questions about what's happening in plain English. Detect events, monitor conditions, and get periodic summaries, all without writing ML pipelines. Powered by Trio's Reality-as-an-API.

When to Use

• - User asks what's happening on a camera, stream, or video feed ("is anyone at my front door?")
• User wants smart alerts for specific events ("tell me when a package is delivered", "alert me if my dog gets on the couch")
• User wants to monitor something they can't watch themselves (construction site, parking spot, warehouse)
• User wants periodic summaries of a live feed ("summarize this stream every 10 minutes")
• User provides any live stream URL: YouTube Live, Twitch, RTSP/RTSPS cameras, HLS streams

Prerequisites

• - A Trio API key. Get one free (100 credits) at https://console.machinefi.com
• Set the key: INLINECODE0
• Base URL: INLINECODE1

Available Actions

1. Check Once (Quick Snapshot)

Ask a yes/no question about what's currently visible on a stream. Costs 1 credit ($0.01).

CODEBLOCK0

Optional parameters:

• - "include_frame": true — returns the analyzed frame as base64 image
• INLINECODE3 — analyze a short video clip instead of a single frame (better for motion detection)
• INLINECODE4 — clip length (1-10 seconds, only with clip/hybrid mode)

Response fields:

• - triggered (boolean) — whether the condition matched
• INLINECODE6 (string) — VLM's reasoning about what it sees
• INLINECODE7 — processing time in milliseconds

Input mode guidance:

• - Use "frames" (default) for static objects: "Is there a car in the driveway?", "Is the door open?"
• Use "clip" for motion/actions: "Is someone walking?", "Did a package get delivered?"
• Use "hybrid" for maximum accuracy (costs more)

2. Live Monitor (Continuous Event Detection)

Monitor a stream continuously and get alerted when a condition becomes true. Costs 2 credits/min ($0.02/min).

CODEBLOCK1

Optional parameters:

• - "webhook_url": "https://your-server.com/webhook" — receive HTTP POST notifications on trigger
• INLINECODE12 — check frequency (5-300 seconds)
• INLINECODE13 — how long to monitor (min 5 seconds)
• INLINECODE14 — minimum seconds between triggers
• INLINECODE15 — set to null for unlimited triggers
• INLINECODE16 — default for live-monitor, good for motion

Response: Returns a job_id. Use it to check status or cancel.

Delivery modes (automatic based on request):

• - If webhook_url is set → events POST to your webhook
• If Accept: text/event-stream header is set (no webhook) → SSE stream
• Otherwise → poll with INLINECODE20

3. Live Digest (Periodic Summaries)

Get narrative summaries of what's happening on a stream at regular intervals. Costs 2 credits/min ($0.02/min).

CODEBLOCK2

Optional parameters:

• - "window_minutes": 10 — summary window length (1-60 minutes)
• INLINECODE22 — frame capture frequency (10-300 seconds)
• INLINECODE23 — receive summaries via webhook
• INLINECODE24 — number of summary windows before stopping
• INLINECODE25 — embed frames in summaries

Response: Returns a job_id.

4. Check Job Status

CODEBLOCK3

Job statuses: pending, running, stopped, completed, INLINECODE31

5. List All Jobs

CODEBLOCK4

Optional query params: status=running, type=live-monitor, limit=20, INLINECODE35

6. Cancel a Job

CODEBLOCK5

Recommended Workflows

Quick Check Workflow

1. 1. Run check-once with the user's question and stream URL
2. Report the triggered result and explanation to the user
3. If the API returns an error about the stream, show the error and remediation

Monitoring Workflow

1. 1. Test the condition with check-once first to verify it works
2. If the condition works, start live-monitor with appropriate settings
3. Return the jobid and inform user how to check status or cancel
4. If webhookurl is available, set it up for push notifications

Summary Workflow

1. 1. Start live-digest with the stream URL and appropriate window/interval
2. Return the job_id so the user can check results later

Condition Writing Tips

• - Frame as binary yes/no questions: "Is there a person visible in the frame?"
• Be specific: "Is there smoke rising from the building roof?" not "Is there smoke?"
• One intent per condition — don't combine multiple checks
• Use positive phrasing: "Are vehicles visible?" not "Is the parking lot not empty?"
• Always test conditions with check-once before starting live-monitor

Error Handling

All errors return this structure:
CODEBLOCK6

Common error codes:

• - NOT_LIVESTREAM — URL is not a live stream. Confirm it's actively broadcasting.
• INLINECODE39 — Cannot reach the stream. Check URL and network.
• INLINECODE40 — Stream exists but is offline. Wait for it to go live.
• INLINECODE41 — Too many concurrent jobs. Cancel old ones with DELETE /jobs/{id}.

If you get an error, always show the remediation field to the user — it contains actionable guidance.

Pricing Reference

Free tier: 100 credits ($1.00) on signup at https://console.machinefi.com

Rules

• - NEVER expose or log the $TRIOAPIKEY value in output shown to the user.
• ALWAYS show the explanation field from check-once responses — it provides the VLM's reasoning.
• ALWAYS test conditions with check-once before starting a live-monitor job.
• When a user provides a stream URL, auto-detect whether they want a quick check, monitoring, or digest based on their intent.
• For monitoring jobs, always return the job_id so the user can check status or cancel later.
• If the API returns an error, show the error code and remediation to the user.
• Inform users about credit costs before starting live-monitor or live-digest jobs (they charge per minute).