Python

Why Python for Agents

What Python is, why agents are built in it, how print and comments work, and how to run code on this site from zero.

Variables and Names

Assignment gives a value a name. Learn naming rules, snake_case, rebinding, and the names every agent loop uses.

Types: Numbers, Text, True/False, and None

Every value has a type. Learn int, float, str, bool, and None, how to convert, and why bool('False') is True.

Compare and Combine Values

Use ==, !=, <, in, is, and, or, not. Know same value versus same object, short-circuit, and beginner traps.

Strings (Text)

Make, slice, and clean text with concat, split, join, strip, and replace — the work of prompts and tool names.

Numbers and Math

Integers, floats, division, rounding, remainders, and how agents count steps, tokens, and cost without lying.

True, False, and None

Truth tests, empty values, None vs 0 vs empty text, and why if result can lie when result is 0 or a blank string.

Lists

A list is a row of values you can change. Index, slice, append, pop, and keep an agent's step memory without losing order.

Index and Slice

Count from 0. A slice is a piece of a list or a string. Off-by-one errors break chunks of text and log windows.

Tuples and Sets

A tuple is a pair you should not change. A set holds unique names, like allowed tools. Empty set is set(), not {}.

Dictionaries

A dict maps a key to a value. Learn [], .get, in, items, nested args, and why this is how JSON and tool calls look.

Nested Data (Traces)

Walk traces as dicts of lists of dicts. Change one field. Chain .get so missing keys do not crash an agent loop.

If, Elif, and Else

Branch with if, elif, and else. Return early. Guard unknown tools, empty text, and errors before the happy path.

Match and Case

match picks the first shape that fits. Route a tool dict with case. Keep if for simple yes/no tests.

Loops and Budgets

for and while repeat work. enumerate, zip, break, continue, any/all, and a hard step budget so the loop cannot run forever.

Functions

Define a function with def, return a value, write a docstring, pass keyword arguments, and store tools in a dict.

Scope, Defaults, and Extra Arguments

Local vs global names, default arguments, *args and **kwargs, and why a list default leaks agent memory.

Unpacking

Split a pair into names, peel the rest of a list, merge dicts with **, and unpack a tool result without losing defaults.

Modules

Import a file of code, rename it with as, pull one name with from, and only run a main block when this file is the program.

JSON and Text

Round-trip Python data with dumps and loads. Know what JSON allows, fake a file with a string, and split a CSV line.

with, Files, and Paths

with always closes. pathlib Path names a file. utf-8 turns text into bytes. This is how read and write tools work.

Errors and try/except

Catch specific errors, clean up with finally, and turn tool failures into dicts the agent loop can read.

Copy vs Change in Place

Lists and dicts change in place. Copy when you must not share. Never give two agents one transcript list.

Comprehensions

Build lists, dicts, and sets in one expression. Filter a trace of dicts. Use a for-loop when the body is more than one step.

Classes

Make a class with __init__, self, and methods. Build a tiny Agent with goal, step, and run. Know when a function is enough.

Dataclasses

Use dataclass to hold data. Freeze a ToolCall snapshot. Set list defaults with field(default_factory=list). Compare with a plain dict.

Type Hints

Write list[str], dict[str, int], and X | None. Hints help you and your editor. They do not stop bad JSON at runtime.

Useful Standard Library

Use datetime, Counter, defaultdict, islice, sha256, uuid, and copy. Know that time.sleep blocks the whole program.

Regular Expressions

Use re.search, findall, and groups. Pull a URL or a small blob from messy model text. Prefer json.loads when you already have JSON.

Logging and Secrets

Log with a list of dicts and levels. Read env vars with os.environ.get. Never print API keys. Redact them with ***.

Virtual Envs and pip

A virtual env is a private folder of packages. Learn pip, pinned requirements.txt, and why Joeven’s browser has no pip.

HTTP and APIs

You ask a server, it answers. Learn methods, status codes, JSON bodies, headers, and timeouts — with fake functions, no network.

Parse JSON from a Model

Models wrap JSON in markdown fences and chat. Strip the fence, parse with json.loads, return an error dict, and never eval().

Retries and Timeouts

Retry timeouts and 429. Do not retry 400 or 401. Cap tries, print backoff waits, and open a circuit after too many fails.

Async in Simple Words

Async means wait for many tools at once. Simulate concurrent jobs with a queue and an in-flight limit. Skip event-loop internals.

Tests

Use assert to test a tool, goal_satisfied, and parse_action. lambda is a tiny check. A runner counts pass and fail.

A Mini Agent in Python

Put it together: a TOOLS dict, JSON parse, a budgeted loop, a transcript, a finish tool, a fake model, and a printed trace.