Google ADK Tutorial: Build Two AI Agents on a Laptop
The tutorial builds a two-agent Google ADK workflow on a 2020 laptop using a free local model through Ollama and LiteLLM. It covers version pins after a compromised LiteLLM release, MCP installation requirements, tool timeouts and the handoff from a researcher agent to a writer.
The demonstration produces a three-line daily plan, taking 55 seconds on a repeat run. A deliberate MCP startup failure exposes fabricated answers, showing why logs matter. The tutorial also explains the limits of ADK's tool-description protections, Gemini API free-tier privacy terms and the Gemini 4 Argon price change on 1 December.
Key points
- ADK requires Python 3.11 or newer, and the tutorial pins ADK and LiteLLM versions after a compromised LiteLLM release.
- The local setup connects ADK to Ollama through LiteLLM and uses MCP tools with a 60-second timeout.
- A Workflow passes the researcher's notes and post titles to a writer that produces a three-line daily plan.
- The local run takes one minute 33 seconds cold and 55 seconds on repeat, but the writer leaves out the post titles.
- MCP tool-description protections are not a guarantee, and a failed server startup lets the model continue and fabricate answers.
- The Gemini API free tier carries privacy caveats, while Gemini 4 Argon is paid and its listed prices rise on 1 December.
Chapters
- 0:00Two local agents
- 0:39ADK versions and installation
- 1:17LiteLLM security and telemetry
- 2:05Ollama and MCP configuration
- 2:49Researcher and writer workflow
- 3:23Local run and results
- 3:59MCP prompt protections
- 4:18Tool failure and fabricated answers
- 4:42Gemini free tier privacy
- 5:07Gemini Argon pricing changes
The description and key points were written with AI from the video's transcript and checked against it, then reviewed by Vishal on . The video plays from YouTube. How we write and check our work.




