Agentic AI on ALIDA
Overview
In ALIDA, the agentic AI layer makes it possible to connect the platform chat to software components that can operate on a real application context, use external capabilities, maintain continuity of work for a specific BDA, and, in some cases, orchestrate other agents or remote servers.
This means that chat is not simply a conversational interface for a model, but a unified access point for agents with different roles. The user experience remains the same, while the underlying agentic backend can change according to the configuration.
What "agent" means in practice
An agent in ALIDA is an operational service exposed through APIs. It must be started on the platform before it can answer calls from the chat or from a workflow. Once running, it exposes a REST invocation endpoint and an SSE endpoint for streaming responses.
The BDA is the application context; the BDAID maintains its identity across the chat, repository, and artifacts. MLflow experiments are named after the parent execution context (FATHER_ID), so runs and traces stay visible from the platform UI: see MLflow Integration. Agent endpoints are REST APIs exposed by the BDA: see I/O Ports: REST API endpoints.
At present, ALIDA's agentic layer centers on two main agent families:
- The ADK Agent acts as an orchestrator and composes external capabilities.
- The OpenCode Agent and Service Writer has an operational role and is specialized for a concrete context through datasets.
First-time setup
For a first-time setup, follow this path:
- First LLM request: verify access to models served through Ollama.
- Choose and prepare an agent: OpenCode or ADK. The agent must be deployed and running, with invocation and streaming endpoints available.
- Configure and use the chat: connect the BDA chat to the running agent endpoint and work in the appropriate context.
- Observability with MLflow: analyze and govern executions.