Phase 1 · Build locally
Build the agent
Eight short tasks on your laptop. You write each step in a starter file and compare it with the reference solution.
0 of 8 tasks complete
How to run a task
uv run bootcamp.py phase1 t2 --starter "How many employees are in Engineering?"
uv run bootcamp.py phase1 t2 "How many employees are in Engineering?"
uv run bootcamp.py phase1 t7phase1/starter/t2_*.py, the second the reference solution phase1/t2_*.py. Add a prompt in quotes for one answer, or leave it out to chat.While a task runs you'll see two kinds of extra lines. [tool] query_db: SELECT ... shows every tool call and its input as it happens (set PHASE1_SHOW_TOOLS=0 to hide them). After the run, [budget] $0.0123 of $1 used (1%), ... shows your LiteLLM spend; it warns at 75% (PHASE1_SHOW_BUDGET=0 hides it). All of Phase 1 usually costs $0.10-0.25 on gpt-6-luna.
Tasks
Setup check
Clone the repo, connect to AWS, and confirm your database copy and model access before Alice's first request lands.
The “Getting Started” challenge
1Basic agent and prompting
Build a polite concierge whose entire knowledge of the company lives in its system prompt.
The “Simple Greeting” problem
2Tools over MCP
Give the agent a key to the company database: a query_db tool served by an MCP server over stdio.
The “Hidden Vault” problem
3File sessions
Persist the conversation to disk so the agent still knows Alice after a restart.
The “Goldfish” problem
4Sliding-window conversation
Control how much history the model sees with a sliding-window conversation manager.
The “Overflowing Context” problem
5Hooks and interrupts
Pause the agent for human approval before any destructive SQL reaches the database.
The “Rogue Agent” problem
6Structured output
Make the agent return a typed Pydantic DepartmentReport instead of a wall of text.
The “Messy Handwriting” problem
7Agents as tools
Split the work: an orchestrator routes to a data specialist and a weather specialist.
The “Overloaded Agent” problem