跟讀練習: 5 open source tools that replaced my $320/mo AI stack... - 通過影片學習英語口語

選擇練習模式

每種模式都會開啟全屏練習室。

63 句

如果句子過短或過長，請點選 Edit 進行調整。

Recently, I did some math, and it made me realize how broken the world is.

A cursor, $20.

Claude Max, $100.

GPT Pro, another $100.

Gemini Ultra, another $100.

Not to mention 11 Labs, DeepSeq, and all the API keys I signed up for two years ago and forgot about.

My AI addiction got so expensive that I had to quit alcohol and Zin.

And I'm not the only one, because alcohol sales in the United States are currently sitting at historic lows.

And Big AI is the one to blame for this.

But that was my wake-up call, and since then, I've cancelled all these subscriptions, and now self-host the ultimate AI stack for developers that's not only cheaper, but far more productive.

In today's video, we'll look at a bunch of free and open-source projects you can run on your own server.

We'll make them work together, while still having the option to tap into the intelligence of big models like Claude and GPT Sol.

It is September 7th, 2026, and you're watching The Code Report.

The first tool we need to talk about is OLAMA, because every AI stack first needs a model.

For most people, this means sending your proprietary code to a trillion dollar corporation and trusting them to keep it totally safe, or you can use OLAMA and do everything locally.

I've talked about this tool many other times on the channel, and essentially it's like Docker for large language models.

It gives you a simple command line interface and API for downloading and running models on your own machine.

This is great if you want to try out the latest Chinese models or any other open weight models, and what's even better is that all your prompts stay private and the inference cost is zero,

which means the model continues to work even after your credit card is declined.

The big problem, though, is that you probably don't own the hardware to run anything close to state-of-the-art.

The small models will run almost anywhere, but running a frontier-sized model would require a small data center, and that's a big problem if you're a Vibe engineer who needs to write top-tier slot.

Luckily, there's a self-hostable solution to this called 9Router, which is an AI tool that sits between your AI tools and dozens of different model providers through one local endpoint.

Instead of juggling nine different API keys, 9Router provides one door with an OpenAI compatible proxy that allows you to point everything through localhost.

But the thing that's really cool about 9Router is fallback tiers.

On tier one, you might have an existing subscription like Claude Max that you already pay for.

Then on tier two, you might have a cheap pay-per-token OpenAI model as a backup.

Then tier three has all of free providers like Chinese models, vertex trial credits, and so on.

That means if you max out Claude Max, it'll automatically roll over to the next tier without you having to do anything.

But in addition to that, 9Router also tracks usage and will compress your tool output to reduce token consumption, which will ultimately save you a bunch of money.

But if you're still eating billions of tokens per day, then you need to know about Headroom, which is a context compression layer for AI agents.

You ask AI to center a div, and to do that, it reads 50,000 lines of your package lock JSON file, then it evaporates an Olympic-sized swimming pool worth of water,

just to realize it needs to install Tailwind CSS.

Headroom sits between your application and the model provider,

and compresses everything from tool outputs to log files and any other useless chunks before it actually goes to the model, which would be billable input tokens.

And one clever design feature of this tool is that it's reversible.

The compressed content that ships to the model is cached on your local machine, which means if the model ever needs it, it can always retrieve it.

This sounds awesome, but at this point, you might be wondering, where do I deploy all this stuff?

Well, the answer, of course, is Hostinger, the sponsor of today's video.

Not only is it one of the most affordable places to host your own virtual private server, but they also have an entire Docker catalog built into the panel,

which means every single open source project I talk about in this video can be hosted there with a single click.

In fact, you could have all these tools running and working together on the same VPS.

Head over to hostinger.com slash fireship and use my coupon code to save a bunch of money spinning up your own VPS.

Once you have that done, the next tool you'll want to install there is Diffy.

Everything we've talked about so far is infrastructure, but now it's time to build an actual app that'll make us rich.

Instead of trying to engineer everything through prompts, Diffy is a visual builder that allows us to drag nodes on a canvas.

This has been a game changer for Horse Tinder and its AI matchmaking feature.

The app will send Diffy each horse's profile information, then a visual workflow retrieves compatible horses from the database, and uses a large language model to explain each match.

Like Thunder and Butternuts are a 94% match because they both enjoy trail rides and biting children.

Then the entire workflow is exposed as an API, so the front-end application can just call it whenever a horse swipes right.

Amazing, but building apps as a human became outdated in Q2 2026 after the release of open hands

which is an open source tool

that allows you to fire yourself this tool is a top performing autonomous coding agent on swebench verified

which is the benchmark where models fix real github issues instead of trying to build

and fix things just open up github issues

and let open hands do all the work essentially it gives

you a command center where you can self-host an army of ai agents

that are always on and working in the background and because Because it's running on your own virtual private server, you have the option to run it with OpenAI or Anthropic models,

or use the local LLMs that you installed earlier with OLAMA.

And now at this point, we have our own private AI stack that could build any type of software you can imagine.

But this is just the tip of the iceberg, and there's tons of other open source applications in Hostinger's Docker catalog that you can install with zero hassle.

Check out the link below

and use coupon code Fireship to get your own virtual private server on Hostinger up and running right now.

This has been the Code Report, thanks for watching, and I will see you in the next one.

下載應用

流利口語所需的一切

AI 發音評分逐句打分

音標練習掌握每個音

詞彙學習建立你的詞庫

詞彙遊戲邊玩邊學

掃碼

Download on theApp Store Get it onGoogle Play

同頻道

FOR YOU

本課的詞彙與口語要點

這節 B2 級別的口語課以影片“5 open source tools that replaced my $320/mo AI stack..”為素材。影片中反覆出現的詞有：model, tool, tier, code, private。這段影片共有 63 個句子、1161 個單詞可供跟讀。講話部分時長為 5:22。說話人語速較快，每分鐘約 217 個詞，會出現較多連讀和弱讀。只有 80% 的單詞屬於英語最常用的 3,000 詞，詞彙難度較高。

影片中的重點詞彙

影片中最難的 15 個單詞，附發音和釋義：

單詞	發音	釋義
tier 名詞	/ˈtɪɚ/	層 /層, 排
token 名詞	/ˈtoʊkən/	代幣 /代幣, 籌碼 /籌碼
compress 動詞	/ˈkɑmpɹɛs/	壓縮 /壓縮
stack 名詞	/stæk/	堆
install 動詞	/ɪnˈstɔl/	安裝 /安裝
hosting 名詞	/ˈhoʊstɪŋ/	寄存服務 /寄存服務
workflow 名詞	/ˈwɝkfloʊ/	工作流
retrieve 動詞	/ɹɪˈtɹiːv/	重拾
coupon 名詞	/ˈkuːˌpɑn/	票證 /票證, 配給券 /配給券
catalogue 名詞	/ˈkæt.əˌlɔɡ/	目錄 /目錄
compatible 形容詞	/kəmˈpæt.ə.bəl/	相容的, 相容的
subscription 名詞	/səbˈskɹɪpʃən/	訂閱 /訂閱, 預訂 /預訂
cursor 名詞	/ˈkɜːsə/	遊標 /游標 or 遊標 /游標, 游標 /游標
endpoint 名詞		端點 /端點
evaporate 動詞	/ɪˈvæp.ə.ɹeɪt/	蒸發掉 /蒸發掉, 揮發 /揮發

影片中出現的短語動詞

單詞	發音	釋義
check out 動詞		退房
try out 動詞		試用 /試用
wake up 動詞	/ˌweɪk ˈʌp/	醒, 醒來 /醒來

影片中的語法

說話人最常用的結構，並附上影片中的原話：

結構	影片中的用法
現在完成時 have/has + 過去分詞 — 過去發生但與現在仍有關聯的事	I've cancelled · I've talked · we've talked
被動語態 be + 過去分詞 — 強調發生了什麼，而不是誰做的	is declined · is cached · can be hosted
定語從句 who / which + 從句 — 補充說明人或事物	consumption, which will · Headroom, which is · machine, which means

需要注意的發音

說話人用了 13 次縮略和弱讀形式，例如 you're, I've, we'll。請按聽到的簡短形式來說。

“sh” 和 “zh” 音: subscription /səbˈskɹɪpʃən/, benchmark /ˈbɛn(t)ʃmɑːk/, cache /kæʃ/, compression /kəmˈpɹɛʃ.ən/, addiction /əˈdɪkˌʃən/
長單詞——注意重音位置: compatible /kəmˈpæt.ə.bəl/, evaporate /ɪˈvæp.ə.ɹeɪt/, reversible /ɹəˈvɜː(ɹ)sɪbəl/, proprietary /pɹəˈpɹaɪ.ə.tə.ɹi/, autonomous /ɔˈtɑnəməs/

中文母語者容易讀錯的音：

/v/ — 不要讀成 /w/，上齒輕觸下唇: provider /pɹəˈvaɪ.də(ɹ)/, retrieve /ɹɪˈtɹiːv/, evaporate /ɪˈvæp.ə.ɹeɪt/, reversible /ɹəˈvɜː(ɹ)sɪbəl/, vertex /ˈvɝˌtɛks/
詞尾子音 — 要讀清楚，後面不要加母音: compress /ˈkɑmpɹɛs/, stack /stæk/, retrieve /ɹɪˈtɹiːv/, prompt /pɹɑmpt/, catalogue /ˈkæt.əˌlɔɡ/
/r/ 和 /l/ 的區別: reversible /ɹəˈvɜː(ɹ)sɪbəl/, builder /ˈbɪl.dəɹ/

如何用這段影片練習

先完整聽一遍影片，不要開口，記下不認識的單詞。
先用 0.75 倍速逐句跟讀，熟練之後再回到正常速度。
錄下自己的聲音並與原聲對比，特別注意 tier, token, compress 這類單詞。

什麼是跟讀法？

跟讀法 (Shadowing) 是一種有科學依據的語言學習技巧，最初開發用於專業口譯員的培訓，並由多語言者Alexander Arguelles博士普及。這個方法簡單而強大：您在聽英語母語原聲的同時立即大聲重複——就像是一個延遲1-2秒緊跟說話者的影子。與被動聽力或語法練習不同，跟讀法強迫您的大腦和口腔肌肉同時處理並模模擬實的講話模式。研究表明它能顯著提高發音準確性，語調，節奏，連讀，聽力理解和口語流利度——使其成為雅思口語備考和真實英語交流最有效的方法之一。

影子跟讀法: 閱讀完整分步指南 →

跟讀練習: 5 open source tools that replaced my $320/mo AI stack... - 通過影片學習英語口語

選擇練習模式

下載應用

The last real programmers are dying out...

PostgreSQL in 100 Seconds

React Native in 100 Seconds

Did Google just dethrone OpenAI and Anthropic?

A $6.3 billion open-weight model just got embarrassed by the French...

Every app on Earth is now open source... whether it likes it or not

OpenAI’s lead engineer was secretly a spy...

Mind-bending new programming language for GPUs just dropped...

Intel chips can’t possibly be this bad… 100% crash rate?

Godot in 100 Seconds

Grok 4 pushes humanity closer to AGI… but there’s a problem

Array Map in 100 Seconds

Docker in 100 Seconds

OpenAI’s new browser feels familiar…

Angular in 100 Seconds

10 weird OSS projects you need right now...

Firefox and Tor hit with 9.8 critical level exploit

The most interesting "hack" in history...

C++ in 100 Seconds

DHH has gone completely off the rails...

How I'd Learn AI From Scratch in 2026 (skip the useless 80%)

10 very promising Open Source Projects you haven’t heard of

An ex-OpenAI researcher just deleted language from the LLM...

Is AI Doom Real... or Just Good Marketing?

AI is flooding the internet—here’s what I’m protecting

Introducing the Agents API

You NEED to try these 12 open-source AI projects RIGHT NOW

Codex Is Winning… So Why Am I Still Paying for Claude?

OpenAI: A Bubble Bigger Than Dotcom

How AI Became More Expensive Than The Workers It Replaced

AI Agents, Clearly Explained

Gemma 4 Just Released - Whats In? In 2026!