Kilo Code vs Augment Code: Which One Is the Best Coding Agent?

Hello again everyone. How's it going?

Welcome back with me Nan. In this video

I want to do a side-by-side comparison

between Kilo Code and Okman code so that

you can figure out which one is the best

coding agent for your workflow. So both

Kilo Code and Okman code are AI coding

agents designed to supercharge your

workflow right inside VS Code. They can

help you write, refactor, and test

complex code with natural language

prompts. Now, on the surface, they might

seem pretty similar, but when you start

using them day-to-day, you'll notice

some key differences. So, in this video

I'm going to compare them side by side

across a few key areas, such as

installation and integration, pricing

performance, and overall user

experience. That way, you'll have a

clear idea of which ones actually worth

using for your setup. First, we'll talk

about the setup. Both Oakman code and

kilo code are extensions that you can

easily install on popular IDE like VS

code or chat brains. Once you've got

them installed, you'll need to sign up

for an account and after that you'll be

able to use the extensions. For Augman

code though, you'll have one less extra

step and that is to index your codebase.

This step is required so that Okman can

make tailored suggestions and explain

common practice patterns. So just let it

index your codebase. Usually it only

takes a few seconds. On the other hand

codebased indexing is optional in kilo

code, so you can start using it right

away. Now, will performance be affected

if you skip indexing in kiloode? Well

we're going to learn more about that in

the performance test. But for now, let's

move on to the interface. So, here's the

chat box for offman code. If you have

used coding agents like GitHub copilot

before, you'll notice that the interface

is very similar. You can send prompts

here or toggle auto approval and then

activate ask mode to ask questions

without changing the code base. And then

you can also select the model to use

over here. The models currently

available on augment code are set 4.5

set 4, GPT5, and haiku 4.5. Augment code

support fewer models when compared to

killer code because originally augment

code doesn't even want to add model

selector to the extension. But as time

goes on and it changes the pricing to

credit based plans, more affordable

models are added so users can save those

credits. Next, Okman can attach image

enhance and select specific context as

well. And now the indexing is also

finished here. H we can see it describes

the project's key characteristics and

features. And then it proposed some

questions that you might want to ask

about. But we'll stop right here and

move on to explore kilo code setup next.

First, kilo code support literally every

AI models that are publicly available.

When you sign up with kilo code, it

might default to something like satan

4.5. But if you open the model selector

you'll find about 500 different models.

Anything from Gemini, GBT, claw to

Deepseek, GLM, Mistral, and so on. Any

model you want to try, simply search for

it, and it will show up here. The UI for

Kilo code is also very similar to GitHub

copilot. Uh you can send prompts and

then use the coding agent to make

changes for you. But what's unique here

is that Kilo code also has five

different interaction modes for its

agent. Each of these modes will define

Kilo code's behavior and determines how

it handles requests. The default is code

mode where killer code acts like a skill

software engineer. It's best for writing

code, adding features or fixing simple

bugs. Then there is architect mode which

turns Kilo code into more of a technical

planner where it can design systems and

lay out implementation plans. There is

also ask mode which simply answers

questions without touching the codebase.

Debug mode that helps you run tasks and

troubleshoot issues. And then

orchestrator mode, an advanced mode for

breaking down and solving complex tasks.

Again, Kilo code has the codebased

indexing feature, but it's optional and

most of the time you don't need it as it

searches for the right context

automatically. And then you can also add

context here adding any folder images or

get comments. And then you can also

enhance with this button. Before we test

the performance of these extensions

let's quickly talk about the pricing.

Honestly, this is one of the things I

really like about Kilo Code because the

pricing basically depends on whatever

API model you decide to use. If you go

with Gemini 2.5 Pro, you'll pay Gemini

API rates. If you choose DeepS, then you

follow Deep Seek's pricing. Kilo Code

also supports third party providers

which opens up some great options, like

the GLM coding plan, which starts at

just $3 a month. And the best part is

you only pay for the killer code

extension itself if you need team

features or want to scale up to an

enterprise plan. If you're on the free

plan, you're really just paying for the

AI model you use and that's it. Okman

code on the other hand starts at $20 per

month which grants you 40,000 monthly

credits. If you need more, then you can

get the higher plan or add 24,000

credits for $15. Now, you probably

wonder how much you can get for 40,000

credits. Well, the answer is it depends

on your usage. Most of the time though

credit based plans are really hard to

figure out. There is a calculation by

one user on Reddit here and as Augment

Code users did protest sharply when they

announced the new plan, it's safe to say

that it's really much more expensive

than before. Kilo Code however believe

transparent usage based pricing is the

way to go. If you use Gemini, you pay

for Gemini. If you use GBD5, then you

pay for GBD5. There is no hidden markups

and no extra fees. Also, when there is a

new model released, Kilo Code usually

partners with the provider and offer the

model for free of charge, even if it's

only for a limited time. When I record

this video, the Miniax M2 model is

currently free in kilo code, as we can

see here. There is also the Gro Code

Fast One model, which is also free to

use at the moment. And when there is a

stealth model being released for public

testing, you can usually try it for free

like the supernova model here. However

note that this free models make use your

data to improve the model. All right

it's time to do a little test. I'm not

going to do a deep test in this video as

it will take the whole day. On the

screen here, I have the code base for a

forum which is a popular open-source

community and social network website.

You can learn more about it at its

GitHub repo here. But the forum codebase

basically has over 6,000 files and 1.7

GB in size. So suppose I'm a new

developer working on this codebase and

I've been given a task to work on the

profile picture upload system. I don't

know where in this codebase to look. So

I'm going to toggle the ask mode and

then send a simple prompt. Find code

that handles when user uploads their

profile picture. All right. Augment code

then starts working on the request here

and then after a while it will start

reading over relevant files and now it

understands the profile picture upload

flow and describe the process back to

us., All right,, let's, scroll, up, a, bit., At

the start here we have the front end

component showing the code for handling

image upload and then there is image

upload action the JavaScript code for

sending the HTTP request. After that

there's the backend controller in Ruby

that process the image and then there's

also an alternative endpoint here. And

then the core business logic is in the

service layer. Um, it seems the chat

scroll is a bit buggy here. It keeps

returning to the previous section. But

for now, uh, we can see the next step in

the flow which is the model layer here

where the carrier wave library is used

to process the file uploaded and then

the uploader class which provides active

data stripping and so on. There's also

key features here such as validation.

Maximum image size is 8 MGB as well as

allowed extensions and then security

processing side effects and then the

flow in a nutshell. It's very

informative and definitely a great

result from augment code. Now let's try

the same task with kilo code. I will

change the mode to ask first and then I

will put the prompt here and click

submit. Let kilo code process the

request. After a moment, it will start

with searching for the right context and

then it will start reading over relevant

files very similar to augment code

before and then killer code still search

for relevant files here and then it

reads the files and after a while it has

found the profile picture upload

implementation in forum. Let it write

the description first and then we will

scroll over from the top. Okay, Kilo

code has finished responding here. Uh

let's go over its report first. Kilo

code starts with the model layer over

here whereas augment code starts from

the front end layer. So the model layer

uses carrier way for handling uploads

and then there is the uploader class

that handles the actual file upload and

then the controller which delegates to

the surface layer. By the way we can

click on this link to open the file in

the editor. Uh the same can be done with

amen as well. Just click on the file

name here. But next let's continue with

the kill code report. So here it

describes the front end part such as the

forms created with embedded Ruby and

then the JavaScript code for handling

the image upload. It ends with a summary

of the flow describing the journey from

user selects the file to updating the

record. So yeah, both extensions are

pretty fast and it can describe the flow

in a way that is easy to understand.

There are subtle differences on the

output such as kilo code here has a bit

more detail about fetching user picture

from social media and then augment

provides the complete process in a

nutshell. But overall we can understand

and start working on the profile picture

upload system with the information

provided by both extensions.

Next let's compare the overall user

experience. If we look at Kilo code here

there are little features and details

that actually make it very convenient.

First, Kilo code has lots of

transparency. You can see at the top

here, there is the cost incurred by the

session so far. And then there are bar

charts over here. Uh those aren't just

for show as they actually reveal the

steps taken by killer code to fulfill

your request. The gray bars indicate

when AI reasons and response and then

the green ones for checkpoint and

completion. The pink one here is for

tool usage. And you can click on any of

the bars to scroll to that step. Next

you can also click on the task to expand

it and then see more details. There is

the context line calculation, how much

is used for the session and how much is

left. And then there is the token usage

showing the amount of tokens sent to the

AI model, how many we received back and

how many tokens are cached. This

information allows you to calculate and

see whether the cost is really accurate.

Now, calculating tokens are quite

confusing. So, I tried to use chat chip

and it seems the result is correct. You

can try to calculate the tokens here

yourself if you want to. And then for

options, Kilo code really has lots of

customization options. You can add MCB

servers through the MCP marketplace over

here. Just search for the MCP and then

click install to add it to Kilo code and

then if you're using many different

model providers, then you can create API

profiles for each provider. For example

here I have the GLM coding plan profile

for when I want to use it. Next, you can

also add more interaction modes as Kilo

code allows you to freely create your

own custom modes. There is also the mode

marketplace where you can install modes

from the community. So, while Kilo code

just works out of the box, it has

optional customizations that can really

improve your workflow. For augment code

here, it's actually a bit more basic as

it has tabs to look for the tasses and

then this one for changes made by the

agent. And then in the auction panel

you can set up MCP and integration

services and so on. And in the chat

session itself, you can't see the cost

incurred by the session or the steps

taken by the AI model. If you want to

look at specific parts in the

conversation, you need to scroll over

the chat yourself to find it. And that's

it. There's really not much else to

explore here.

All right. Now that we've compared how

Kilo code and augment code works, it's

time to decide which one is the best

coding agent. Now, this is going to be

subjective, but I do prefer Kilo code

over Augment Code. I mean, sure, Augment

Code might be a little easier to get

started with, but that simplicity comes

with trade-offs, like higher pricing

less transparency, and less control over

the configurations. As AI coding tools

move from being just fun experiments to

becoming part of our everyday workflow

developers are going to want more

control and transparency. And if you

work for a company, you'll definitely

realize that managers want clearer

visibility into causes to measure return

on investment. Also, everyone wants

better results from their coding agents

which means more customization to fit

their unique cases. And that's exactly

where Kilo Code shines. Its usage based

pricing means you only pay for what you

actually use. There is no surprise rate

limits and definitely no sudden price

hikes. Kilo Codeoke isn't trying to

profit of how much you interact with the

AI model. It also doesn't force you to

use specific models. Although it does

have some tips and guidelines for

choosing the best model. On top of it

all, Kilo Code is also open source. So

if you want to check out the code, learn

how it's built or help improve it, then

you can visit its GitHub repo. So yeah

both Kilo Code and Oakman Code are solid

tools for serious developers. But if you

care more about control, customization

and transparency, then I would say go

with Kilo Code. All right, that's it for

this video. Hopefully, you have a better

idea of which coding agent fits your

workflow better, whether it's Kilo Code

or Augment Code. I hope you all enjoyed

today's video and get some value out of

it. Let me know your thoughts in the

comments below. I'll join the

conversation and reply as often as I

can. If you're new to the channel, my

name is Nathan and I help you build

profitable apps and projects using AI

and other tools. Make sure to subscribe

if that's something you find useful.

Don't forget to like this video, turn on

the notification bell, all that good

stuff, as it really helps the channel to

grow. With that being said, thanks so

much for watching until the end. I hope

you have a great day and I'll see you in

other videos. Bye-bye.

[Music]