Dialect
Developer Tool
Lets AI agents read the physical world through cameras. An agent proposes typed fields from regions of the frame, a person confirms them, and a Raspberry Pi gateway keeps them current over REST and MCP.
Python
FastAPI
PostgreSQL
Raspberry Pi
MCP
AWS
Terraform
Key features
- An agent proposes an endpoint, which is a set of typed fields read from regions of the camera's frame. A field can be a choice, a yes or no, a level, or a light's state
- A person confirms every endpoint before it goes live
- A Raspberry Pi gateway keeps each endpoint current
- Agents and programs read endpoints by name over REST or MCP
- A read-only MCP route serves agents that should change nothing
- Camera-only. It reads what a camera sees and acts on nothing
- The API and PostgreSQL run on AWS (ECS and RDS), set up with Terraform
Dialect gives the physical world an API. A gateway’s camera looks at something real, and Dialect turns what it sees into typed values a program can read: a level, a yes or no, a light’s state.
How it works
- An agent looks through the camera and proposes an endpoint. An endpoint is a set of typed fields, each read from a region of the frame.
- A person confirms the endpoint.
- The gateway, a Raspberry Pi, keeps the endpoint current.
- Anyone with access reads it by name, over REST or MCP.
What it doesn’t do
Version 1 reads and does not act. Earlier versions had an infrared emitter. I removed it, so Dialect now only reports what a camera sees.
The code is private.