ToolExperimentalAG-UI

ebi-agent-chat-relay

Name: ebi-agent-chat-relay
Rating: 3.0 (57 reviews)
Author: ebibibi

by ebibibi

Relay agents to Discord and Teams with ag-ui and LLM adapters

Python

Updated Sep 21, 2026

Stars

Forks

View on GitHub

Overview

Connects multiple frontends (Discord, Microsoft Teams) to agent backends like Claude Code, Codex, and local models. Acts as a relay that forwards messages between chat platforms and AG-UI-compatible agents, normalizing inputs and outputs for multi-source interaction. Includes adapters for different LLM providers and a simple routing layer to run agents from chat clients without embedding model logic in the frontend. AG-UI Protocol and Dynamic Task Routing Pattern help standardize how these routes are managed across frontends.

See related protocols

Standards this tool supports

The Value Proposition

As agents are embedded into everyday chat platforms, observability and consistent handling of interactions become essential for evaluating agent behavior. A relay standardizes how agents are invoked and logged across frontends, making it possible to collect interaction traces and compare agent responses under identical conversational contexts. That consistency is a first step toward agent-to-agent evaluation and tracking agent performance across real user touchpoints. Model Context Protocol (MCP) can guide how context is shared across agents for better traceability. Continuous Monitoring also supports ongoing visibility into agent behavior.

Ideal For

Teams prototyping multi-fronted agents who want a quick way to run Claude/Codex/local agents from Discord or Microsoft Teams. For a aligned approach to context and interactions, consider the Model Context Protocol (MCP) and related patterns to integrate smoothly with AG-UI capable workflows.

Applications

Route user messages from Discord/Teams to different LLM backends for side-by-side comparison
Capture consistent interaction logs across chat frontends for downstream evaluation
Prototype agent delegation patterns by forwarding tasks to specialist models without changing the frontend