Developers
Set up Ray from the terminal or from your own assistant
A JSON definition file that the business’s AI assistant can fill in, and later a CLI and an MCP server to automate knowledge, tests and reports. Always with the same permissions and the same test cases.
Early access with hands-on help from the team. Ray is in development.
Illustrative example of an agent definition file in JSON, in the ray-agent/1 format, with an internal name, persona names, the Kind tone, instructions, two knowledge collections and no tools.
Illustrative example with sample data. Lumora is a fictional brand.
Status
What is in development and what comes next
The definition file will come with the agents. The CLI and the MCP server will use per-business keys, which will come after the credential store and a security review. No committed dates.
- In development
Definition file and brief
Import and export agents as JSON, and a brief for the business's assistant to fill them in.
- Coming soon
CLI
Upload knowledge, run the test cases, export agents and read reports from the terminal.
- Coming soon
MCP server
The business's assistant will query Ray with its own key: reading by default, writing only as a proposal.
Definition fileIn development
The business's assistant will draft the agent; the console will validate it
The console will hand over a brief to paste into the business's AI assistant. The assistant will run a short interview and write the JSON; once imported, it will become a draft agent that goes through the same checks as one built by hand.
- Copy the brief from the console
- Paste it into the business's assistant and answer the interview
- Import the JSON: a draft agent will be created
- Test and publish, like any changeAny agent will be exportable in the same format.
Illustrative example: above, the setup brief with a button to copy it; below, an import report with valid fields, the collections found, no tools assigned and publishing pending, with buttons to open the draft or export as JSON.
Illustrative example with sample data. Lumora is a fictional brand.
CLIComing soon
Ray from the terminal
For teams that keep their knowledge in files or want to run the tests in their own release process. It will install with pnpm and use the business's key.
Upload knowledge
A folder of md, txt or html files, as drafts or new versions to review.Run the test cases
The same set of cases as in the console, with the result per language.Export agents
The same JSON as the definition file, to version it with everything else.Read reports
Content gaps and resolutions, as text or JSON.
Illustrative example of a terminal session: install the CLI with pnpm, sign in with the business key, upload a help folder that leaves unpublished drafts, run the test cases, export an agent to JSON and read the content gaps.
Illustrative example: the commands may change.
MCP serverComing soon
The business's assistant, connected to Ray
An MCP server will let the business's AI assistant, or another compatible assistant, work with Ray using its own key. Reading will be the default; writing, always a proposal a person approves.
- Read-only by default: search the knowledge and read reports.
- Every write will be a proposal, like a draft article, that a person approves.
- Running the test cases will never publish anything.
- It will come after the credential store and the security review.
Illustrative example of the MCP server's tools: search the knowledge and read reports, read-only; propose a draft article, as a proposal; run the test cases, without publishing; and the business's active key with the Editor role, which can be revoked.
Illustrative example: the tool names may change.
Security and control
The same rules from the console, the terminal or an assistant
Keys will be per business and per role, every action will be recorded and no change will skip the test cases.
Per-business keys
Each key will belong to one business and one role, and can be revoked at any time.
Shown only once
After it is created, the key will not be shown in full again: if it is lost, it will be revoked and a new one created.
Audit log
Every action will be recorded with the key, the person who created it and the result.
The same test cases
A change that comes through the CLI or MCP will go through the same cases and the same sensitive-topics floor.
Talk about automated setup with the Ray team
A short conversation with the team that builds Ray to review the company’s case: frequent questions, existing knowledge, sensitive topics and a migration plan.