Assistant ERI Server

Deutsch: Assistent ERI-Server

The "ERI Server" assistant (beta) generates the code of an ERI server from your description. ERI, the External Retrieval Interface, connects AI Studio to the data of your organization: you run an ERI server next to your data, and AI Studio uses it as a data source; see Data Sources. The assistant is meant for developers and IT departments who build such a server.

Switching it on

The assistant belongs to the data sources, a beta feature. In "Settings" → "App Options":

1. Under "Preview feature visibility", choose "Also show features in beta: these are almost ready for release; expect some bugs".
2. Under "Select preview features", choose "RAG: Preview of our RAG implementation where you can refer your files or integrate enterprise data within your company".

Its tile then appears under "Software Engineering".

Presets and auto-save

Under "ERI server presets", you keep one configuration for each ERI server you are responsible for. "Add ERI server preset" adds one; click it in the list to work on it. "Delete this server preset" removes the selected one after asking you.

With "Should we automatically save any input made?" switched on, the assistant saves every change to the selected preset. Presets and auto-save need "Preselect ERI server options?" in the options of the assistant, which is on from the start.

Describing your server

Work through the sections of the form:

• "Common ERI server settings":
  • "ERI server name" (6 to 60 characters) and "ERI server description" (32 to 512 characters). Users see both in AI Studio, so say what data the server provides and what for.
  • "Programming language", for example, C#, Java, Python, or Rust. "Other" lets you enter any language.
  • "ERI specification version": choose V1. "Download specification" opens the specification.
• "Data source settings": where your data lives, for example, "File system (local or network share)", "Relational database, like MySQL, PostgreSQL, etc.", or "Custom description". For a database or a storage, you also enter the product name, the hostname, and the port.
• "Authentication settings": "Authentication method(s)", from no login for public data to a login by token, by username and password, or by single sign-on with Kerberos. With several methods, describe how they are meant to be used. For Kerberos, you also choose the operating system of the server.
• "Data protection settings": "Allowed LLM providers for this data source", any provider or only self-hosted ones. The server reports this choice to AI Studio, which then gives its data only to such providers.
• "Embedding settings" (optional): "Add Embedding Method" describes an embedding method your server uses: name, type, description, and when it is used.
• "Data retrieval settings": "Add Retrieval Process" describes how your server finds the passages that answer a question, with optional parameters and the embedding methods it uses. Describe at least one; with several, users can choose.
• "(Optional) Additional libraries": libraries the code should use, and what for. Models may make up libraries that don't exist, so check them.

Generating the server

1. Under "Provider selection for generation", choose a provider. The task needs a very powerful model; small local models will probably not be enough.
2. Under "Write code to file system", decide whether AI Studio writes the code to disk. If so, choose the "Base directory where to write the code".
3. Click "Create the ERI server".

The model first lists the files of the server, then writes them one at a time, and finally explains how to compile and start the server. This takes a while and many requests; AI Studio respects the rate limits of your provider by itself. The whole conversation appears on the page. "Open in chat" continues it in a chat, in the workspace "ERI Servers". "Copy result" copies only the last answer.

Writing to disk

AI Studio writes only inside the base directory. When you generate again, it first deletes the files it wrote last time; your own files stay, unless the model writes a file with the same name. It doesn't ask before it writes or deletes. Therefore, keep the base directory in a Git repository, and commit or revert all changes before each run.

Options

In "Assistant: ERI Server Options", "Preselect ERI server options?" enables presets and auto-save, and lets you choose a minimum confidence and a profile. The other settings live in the presets of the assistant.

Tips

• The generated code is a starting point. Review it before it handles real data, above all the login and the data protection.
• Compare the files on disk with the file list in the answer, and add what is missing by hand or with the Coding assistant.
• The ERI repository contains a reference implementation in C#; see Developers.

Where to go next

• Assistants
• Data Sources: connect AI Studio to your ERI server.