Gemini CLI and Code Assist
Professional developers who want Google's models as an engineering tool: in the terminal, in the IDE, and behind their own code
Prerequisites: A working developer setup (terminal, git, an IDE, Python or Node). Module 18 (Prompt Engineering Deep Dive) is assumed. Module 201 is not required.
Last verified 2026-09-17
Learning objectives
By the end, participants can:
- Map the surfaces to the accounts that can use them (since 18 June 2026 the free and Google AI Pro and Ultra tiers run Antigravity CLI (agy) instead; Gemini CLI continues for Gemini Code Assist Standard and Enterprise licences, paid Gemini API keys and Vertex AI) and pick the right one for their situation
- Install and authenticate Gemini CLI (npx @google/gemini-cli, npm install -g @google/gemini-cli, or brew install gemini-cli; Apache 2.0) or Antigravity CLI, and know which credential they are using
- Write a
GEMINI.mdthat earns its place (GEMINI.md, loaded from ~/.gemini/GEMINI.md, the workspace and its parent directories, and just-in-time from directories tools touch; /memory show prints the combined result and /memory reload rescans; @file.md imports; rename via context.fileName in settings.json) - Configure an MCP server and set its trust deliberately (mcpServers in settings.json, with stdio (command and args), SSE (url) or streamable HTTP (httpUrl) transports; trust is false by default so tool calls ask first; /mcp lists servers and tools)
- Use Code Assist agent mode in the IDE and say how it differs from the terminal
- Get an API key safely (created in Google AI Studio; new keys are auth keys bound to a service account, and legacy standard keys are rejected from September 2026; read from the GEMINI_API_KEY environment variable), call a model (pip install -U google-genai, then client.interactions.create(model=..., input=...) in Python), declare a tool, and run the function-calling loop
- Write a first agent loop and place the autonomy line in it
What's in this module
7 lessons · 7 exercises · ~10 min reading
Starts withThe map, and what changed in June 2026~3 minExecutive summary
You will learn the four ways a developer reaches Gemini, which of them still work for which account after the June 2026 changes, how to run an agent in the terminal with a context file and MCP servers, how the IDE agent mode differs, how to prototype in AI Studio and ship on the API, and how to build a first agent loop of your own.
End state is a terminal agent authenticated in a way that will keep working, a committed GEMINI.md in a real repository, one MCP server connected, a working API call with a tool, and a small agent loop you wrote and can explain.
After this module you can choose between Gemini CLI, Antigravity, Code Assist and the raw API for a given job, write a context file that changes behaviour, connect and trust a tool, draw the autonomy line for your team, and build the loop that everything else is a harness around.
Sources
- Gemini CLI repository, the README: what it is, install routes, built-in tools, the Apache 2.0 licence: https://github.com/google-gemini/gemini-cli
- Gemini CLI documentation, "Authentication", the current sign-in options and the June 2026 notice on which tiers moved to Antigravity CLI: https://geminicli.com/docs/get-started/authentication/
- Gemini CLI repository, "Provide context with GEMINI.md files", the hierarchy, imports and the memory commands: https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/gemini-md.md
- Google Cloud, "Gemini Code Assist Standard and Enterprise overview", the editions, supported IDEs and what Enterprise adds: https://docs.cloud.google.com/gemini/docs/codeassist/overview
- Google AI for Developers, "Function calling with the Gemini API", tool declaration, the loop, and remote MCP as a tool type: https://ai.google.dev/gemini-api/docs/function-calling
- Google Antigravity, "Migrating from Gemini CLI", what carries over and what moved: https://antigravity.google/docs/cli/gcli-migration/
Prefer the live room?
This module also runs inside our in-person bootcamps and workshops in Copenhagen.
More in Track F
Prompt Engineering Deep Dive
The complete prompting toolkit, from structure and roles to hallucination control and complex industry prompts, plus how to measure prompt quality with evaluations.
Claude Code for Developers
How to run Claude Code as a professional engineering tool: operating manuals, subagents, hooks, skills, MCP, and git/CI workflows.
Agent Harness Engineering
How agents actually work under the hood: the model supplies intelligence, and everything else is a harness you can build.