Cody

On-device AI, sports, gaming, software dev. I make esper, it’s like ChatGPT but it runs 100% offline on your iphone. Download ⬇️ Website ➡️

[Cody @keuxdi]

These images were generated on an iPhone in less than 10 seconds 🤯

[Cody @keuxdi]

Everything you need.

[Cody @keuxdi]

You can run AI locally. You can use privacy focused cloud providers with TEEs. You can redact sensitive data and wipe memory.

But every privacy solution today requires more effort than just opening ChatGPT.

I hope once model intelligence plateaus, the focus shifts to privacy.

[Cody @keuxdi]

There’s no reason for chat clients to lock you into one model per conversation.

GPT 5.5 ➡️ Apple Foundation Model

[Cody @keuxdi]

🫩

[spor @sporadica]

Maybe it’s the Gen-Z in me, but i fully don’t care about privacy. I am post-privacy. I am giving OpenAI access to all of my finances, all of my health data, everything, I don’t care anymore

[Cody @keuxdi]

Running Gemma 4 E4B on my mouse

[Eli @elkelk]

fun fact- the new MX Master 4 has 16gb of ram to manage the cross device smart clipboard

its $299.99 because of this

[Cody @keuxdi]

When Apple finally rebrands Siri around modern AI, they should revive the “Shot on iPhone” campaign and turn it into “Running on iPhone.”

[Cody @keuxdi]

🚫🛜

RAG working locally on your phone.

[Cody @keuxdi]

Freedom of thought 🫡

[Mark @marksuman]

In order to have freedom of thought in society, we need tools that allow us to learn, research, and ask questions. A system that tracks its users and enforces an opaque ruleset does not promote freedom.

We built @TryMapleAI to support freedom of thought.

I don't have the ability to tell you what is the objective truth on every subject, so I'm not going to impose my view on our users.

The best path forward is open, verifiable AI. Open models, open code, and either local hardware or TEEs.

[Cody @keuxdi]

You can get unlimited image generations by running the model on your phone 💭

[Cody @keuxdi]

Most phone apps that run local models are really just gimmicks and demos at this point.

But once you add RAG, voice, vision, web search etc. it starts to become a useful tool.

[Cody @keuxdi]

People don’t know iOS has a built in LLM you can talk to offline and free

[Cody @keuxdi]

Relevant:

[Cody @keuxdi]

@gabriberton RAG, summarization, translation, journaling, companions, text or image classification, q&a with web search.

Privacy in general is one of the biggest upsides overall.

[Cody @keuxdi]

Local AI is not worse than cloud AI. It’s just a different tool.

[Cody @keuxdi]

Privacy filter is super cool tech, I plan on implementing it as a pre filter for API provider calls in esper.

But the truth is as soon as any prompt or file leaves your device, you have absolutely no guarantee on how it will be used.

Always use the right tool for the job.

[clem 🤗 @ClementDelangue]

OpenAI dropped a new model on HF today!

[Cody @keuxdi]

@gabriberton RAG, summarization, translation, journaling, companions, text or image classification, q&a with web search.

Privacy in general is one of the biggest upsides overall.

[Cody @keuxdi]

Some of my favourite features coming soon:

🌆 Local image generation with stable diffusion

🔎 Web search with local models

📁Media/file gallery

🤬 Abliterated Gemma 4 variants

[Cody @keuxdi]

@_amankishore Love more people discovering use cases for llms on phones

[Cody @keuxdi]

@stonej09 Yeah np

[Cody @keuxdi]

I've been hesitant to add an obliterated model to esper... maybe it's time

[Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 @elder_plinius]

😱 HOLY SHIT... Someone just dropped a fully liberated Gemma 4 E4B!

and the guardrail removal process appears to have left coherence fully intact AND improved coding abilities! 🤯

https://t.co/XeednUqsrM

OBLITERATED Gemma:
✅ 97.5% compliance rate, 2.1% refusal rate, 0.4% degenerate outputs
(499/512 prompts answered on OBLITERATUS bench)

ORIGINAL Gemma 4 E4B:
❌ 1.2% compliance rate, 98.8% refusal rate
(506/512 prompts refused)

Coherence: fully intact
Factual: same
Reasoning: same
Code: +20% 📈
Creative writing: same

But the REAL story here isn't the model itself, it's how it was made...

🧵 THREAD 👇

[Cody @keuxdi]

@stonej09 I’ll most likely add it in the next update. App is free to try!