Qlaudia

← All courses

Build an Agentic AI App in Elixir

  • Artificial Intelligence
  • Technology & Computing

Elixir agents combine language models with tools and feedback loops to reason and act autonomously, moving beyond single-shot responses to iterative problem-solving. Build a research assistant that searches the web and private knowledge bases, weighs conflicting sources, and serves answers via terminal or web UI.

What you'll learn

  • ContentsNavigational table of contents for the tutorial.
  • Step 3 — Module 1: a model, a tool, and the loopBuilding the core Model, Tool, and Step abstractions, a first web-search tool, and the THINK-ACT-OBSERVE loop into a working ToolCallingAgent script.
  • Step 11 — The capstoneCombining all prior modules into a capstone research-assistant agent with CLI and web entry points.
  • Step 5 — Module 3: a second agent style that writes codeBuilding a CodeAgent that has the model write and evaluate Elixir code directly, with sandboxing.
  • Step 2 — Dependencies, config, and your API keyConfiguring mix.exs, the Application module, runtime config, and environment files, then fetching dependencies.
  • What you'll buildOverview of the finished research-assistant agent and its capabilities.
  • Before you startSetup expectations, what the Anthropic API key is for, and what to expect from the tutorial process.
  • Step 8 — Module 6: loading tools over MCPBuilding an MCP server and client from scratch to load tools from another process.
  • Step 6 — Module 4: retrieval (RAG)Adding local embedding dependencies and building a knowledge-base tool with cosine similarity retrieval.
  • Step 9 — Module 7: a streaming web UIBuilding a Plug/Bandit web UI with a router and streaming responses via Server-Sent Events.
  • What is an agent?Explains what an agentic AI system is and how tool calling actually works under the hood.
  • A note on safety / Where to go nextClosing remarks on safety considerations and pointers for further exploration.
  • Step 7 — Module 5: multiple agentsImplementing a manager/worker multi-agent pattern using as_tool/1 and a page-reading tool.
  • Step 10 — Module 8: seeing what it did and what it costAdding telemetry helpers for observability and per-run cost tracking.
  • Who this is for / What you'll needIntroductory notes on the target audience and prerequisites for the tutorial.
  • Step 4 — Module 2: writing your own toolAdding a hand-written weather tool to the agent and verifying it works.
  • Step 1 — Create the Mix projectCreating the new Mix project and setting up .gitignore.