Give the physical world an API.

Describe what a camera should observe. Get a live API your software and agents can use. Nothing to train.

Create account
Endpoints
Desk valve
Looking at the camera
DeclineStart reading
Live10:17:44 PM
ValveShut
Tell me when the valve opens.
dialect_setup
I’ve set up Desk valve. Start reading it and I’ll watch.
It just opened.

Stop walking over to check

Real footage, from a hangar to a reservoir, each set up from the question beside it.

Hangar bay
Live
Bay freeYes
Is the hangar bay free?
read_bay_free
Yes, the bay is empty.
Is the hangar bay free?
read_bay_free
No. An aircraft was towed in a moment ago.
Is the hangar bay free?
read_bay_free
Yes, it’s free again. The aircraft was just towed out.

Use the cameras you already have

Any camera that speaks RTSP or ONVIF. Copy one prompt to your agent. It finds the camera and starts Dialect on a computer beside it, and you click Add.

Gateways
Enrolled and invited5
Add a gateway
GatewayStatusEndpoints
loading-dockOnline0
cold-roomHealthy3
yard-gateHealthy2
workshopHealthy6
north-lotHealthy4
freezer-2Healthy1
Add a gateway
Camera addressrtsp://user:password@192.168.1.20:554/stream1
Copy commandCopy for your agent
Connect a camera to Dialect as a gateway, from a shell on a computer that’s always on and on the camera’s network…
Looking for cameras
Found the dock camera and started Dialect next to it. Open this to add it:
dialectlayer.com/claim/3f9c2e7a41d0
Add this server gateway
It asks to be called loading-dock.
Add to Kestrel Logistics

One camera, many questions

Every question you ask is its own endpoint on the same camera, set up from a sentence. Nothing is trained for any of them. Here one camera at NASA’s Webb test chamber answers four.

chamber-a
LiveJun 13, 10:16 AM

What it reads0

Chamber doorIs the chamber door shut and sealed?
No
TelescopeIs the telescope inside the chamber or out on the floor?
Out on the floor
Work platformIs the yellow work platform set up in front of the chamber opening?
No
CraneIs the crane hooked up and holding the telescope?
Yes

Your software acts on it

Read an endpoint from an agent or your code, or have each change pushed to your app. Here the deck clears, the webhook fires, and a sample app clears the next aircraft.

  • MCP

    dialect_read_endpoint

    A tool for Claude, Codex or any MCP client, given the endpoint’s name.

  • REST

    GET /v1/endpoints/{name}

    Every field’s value, how sure it is, and since when it has held.

  • Webhooks

    endpoint.changed

    A signed request to your app each time a value changes, with the value it was.

Flight deck
Live
Aircraft on deckYes
WebhookPOST your-app.com/dialect
Sends each change as it happens
Your code
if (data.value === false) { queue.clearNext()}
Your app
Next aircraftWaiting
After itWaiting

Read one now, no account

Chamber A at NASA’s Webb test chamber is a public endpoint. Call it from a terminal, or connect your agent and ask it whether the chamber is sealed.

Chamber A now

Reading…

The camera is a looped public-domain NASA recording of the chamber opening, read live by Dialect as any camera is.

From a terminal

curl https://api.dialectlayer.com/v1/public/chamber-a

From Claude Code

claude mcp add --transport http dialect-demo https://api.dialectlayer.com/mcp/public

From Cursor or any MCP client

Ask it things nobody planned

This endpoint answers one question: whether an aircraft is in the bay. Nobody asked it about the engines. An agent asks anyway, and Dialect answers from the same camera.

Hangar bay
Live
AircraftYes
EnginesOn the wing
Engines3 on stands
Are SOFIA’s engines still on the wing?
dialect_look
Yes. Both on the near wing are mounted.
Are SOFIA’s engines still on the wing?
dialect_look
No. Three are on stands beside it. The fourth is out of view.

Then it keeps watching

Dialect tells you when something changes, keeps the day’s history, and carries on when the camera moves or the network drops. When it can’t tell, it says so instead of guessing.

  1. Desk valve
    Live10:17:44 PM
    ValveShut
    10:17
    Saturday, October 24
    10:17
    Desk valve: Valve is back to ShutnowSeen 10:17:56 PM
    Desk valve: Valve is OpennowSeen 10:17:49 PM
    Desk valve: Valve is back to ShutnowSeen 10:17:41 PM
    Desk valve: Valve is OpennowSeen 10:17:34 PM

    Tell me when it opens

    Ask to be told when the valve opens. The alert reaches your phone a moment after the hand turns it.

  2. Endpoints
    Desk valve
    OpenShut
    3 PM6 PM9 PM
    Live10:17:44 PM
    ValveShut
    How long was the desk valve open today?
    dialect_summary
    4 h 45 min. It was opened twice, at 3:03 PM and 10:17 PM.

    How long was it open today?

    Dialect keeps every value with the time it changed, so an agent can answer about the whole day, not only about now.

  3. Desk valve
    Live10:17:44 PM
    ValveShut

    When it can’t see, it says so

    A hand covers the valve. The reading says Covered instead of guessing, then goes back to Open once the hand is gone.

  4. Desk valve
    Live2:48:20 PM
    ValveOpen

    The camera moves, the reading holds

    The camera hangs from a standing desk. When the desk goes up, Dialect finds the valve again and reads on.

$99 a month. $49 for each question after the first.

Your first answer is free. Then 30 days, full refund.

Create account

A question is one endpoint. It can answer with several fields. A camera answering four questions is $99 + 3 × $49 = $246 a month. Up to 10 on this plan. Every reading, alert, person and agent is included.

Custom

For more than 10 questions or security and contract needs. Talk to us: tyler@grayhavenindustries.com

Common questions

  • What is Dialect?

    Dialect turns a camera into a live API. You describe what it should observe in a sentence and confirm what it proposes. Your software and agents then read typed values, with how sure each one is and since when it has held. Nothing is trained for any endpoint.

  • Which cameras does Dialect work with?

    Any camera that speaks RTSP or ONVIF. Dialect runs in Docker on a computer beside it.

  • Do I have to train a model?

    No. Every endpoint is set up from a sentence, and nothing is trained for it.

  • How do my software and agents read a camera?

    Three ways: an MCP tool for Claude, Codex or any MCP client; a REST call to GET /v1/endpoints/{name}; or a signed webhook each time a value changes.

  • What happens when Dialect can’t tell?

    It says so instead of guessing, and every value says how sure it is.

  • How much does Dialect cost?

    $99 a month. $49 for each question after the first. A camera answering four questions is $99 + 3 × $49 = $246 a month. Your first answer is free. Then 30 days, full refund. Custom is for more than 10 questions or security and contract needs.

Ask your first question

Create an account, connect a camera and ask one question, free. Pay to keep it reading.

Connect your agent

Or ask in the console once you’re in.

claude mcp add --transport http dialect https://api.dialectlayer.com/mcp

Then run /mcp, choose dialect and sign in. It reads Connected. No token needed.

Then ask it

What can Dialect’s cameras see right now?