Gemma 4 + React Native: Run AI Fully Offline In Any App No Cloud, No API
A technical walkthrough of running Google's Gemma 4 edge model fully offline inside a React Native app via the ExecuTorch library, covering the memory math behind its quantized formats and the useLLM hook that handles loading, streaming, and on-device tool calling.
AI Stack Engineer8 minTranscript found
Quick learning frame
Read this before watching.
Creative automation uses agents to accelerate production while keeping human taste in story, pacing, selection, and critique.
New playlist item from AI Stack Engineer; queued for transcript-backed review, topic mapping, and a practical learning artifact.
Skill you build: The ability to reason about a language model's on-device memory footprint (precision, quantization, and size tier) to pick the right model variant for a target mobile device.
Watch for the shift from claim to mechanism. The learning value is the point where the transcript reveals a repeatable action, tool boundary, context move, review habit, or artifact.
Concept diagram
Where this video fits.
01Brief
02Source
03Generation
04Selection
05Edit
06Taste Review
Deep lesson
Turn this video into working knowledge.
1,466 cleaned transcript words reviewed across 436 timed caption segments.
Thesis
Gemma 4 + React Native: Run AI Fully Offline In Any App No Cloud, No API teaches a practical creative automation move: A technical walkthrough of running Google's Gemma 4 edge model fully offline inside a React Native app via the ExecuTorch library, covering the memory math behind its quantized formats and the useLLM hook that handles loading, streaming, and on-device tool calling.
The goal is not to remember the video. The goal is to extract the operating principle, tie it to timestamped evidence, test how far the claim transfers, and make something reusable.
0:32
Apache 2.0 Unlock
“4 support to their library. React Native Executor Torch is an open-source project from a company called Software Mansion, and it lets you run machine learning models directly on the phone using PyTorch's Executor Torch runtime. The repo...”
Gemma 4 ships under Apache 2.0 (unlike earlier Gemma versions' restrictive Google license), in five sizes (E2B, E4B, 12B, 26B, 31B) where 'E' means 'effective'/edge-sized for phones; the 31B dense model ranks in the top three on the Arena AI leaderboard despite being 20-50x smaller than the models it's competing against, and every size supports toggleable step-by-step reasoning, native multimodal input, function calling, up to 256K context, and 140+ languages. Compare Gemma 4's five sizes against your target device's RAM and note which size (E2B, E4B, 12B, 26B, or 31B) is actually viable before you start integration work.
3:02
Quantization Math
“natively, which is the part that lets the model actually do things instead of only chatting, and they hold a lot of context, up to 256K tokens on the larger sizes. They also support over 140 languages, which...”
Google used QAT (quantization-aware training) so the model learns while accounting for compression, letting a mixed 2/4/8-bit 'mobile format' shrink E2B from about 3GB (standard 4-bit) down to roughly 1GB, or under 1GB (about 0.8GB) if you drop the vision and audio encoders for text-only use; larger sizes scale up fast, E4B at ~5GB, 26B at ~15GB, 31B at ~18GB, meaning E2B is the only size that realistically fits on a phone. Calculate the RAM budget of your target phone and work backward to figure out whether you need the text-only (~0.8GB) or full multimodal (~1GB) E2B build.
6:36
On-Device Tool Calling
“an on-device model useful instead of just a chatbot. You define a tool, basically a function with a name, a description, and its parameters. You hand that definition to the model. When the model decides it needs that...”
ExecuTorch's useLLM React hook handles fetching, loading, and streaming a Gemma 4 model token by token, using the Vulcan delegate on Android and MLX on Apple silicon to hit 60+ tokens/sec; tool calling works by handing the model a function's name, description, and parameters, letting it emit a structured call that your code executes locally before returning the result, which is the exact mechanism behind the flyer-to-calendar-event demo, and the whole point is offline reliability, on-device privacy, and no per-token cloud cost. Wire up one real tool (e.g., a calendar-add function) to the useLLM hook's tool-calling interface and test that the model correctly emits a structured call for it on a real (not simulator) older phone.
01
Brief
Start with this video's job: A technical walkthrough of running Google's Gemma 4 edge model fully offline inside a React Native app via the ExecuTorch library, covering the memory math behind its quantized formats and the useLLM hook that handles loading, streaming, and on-device tool calling. Treat "Brief" as the outcome you are trying to make visible, not a topic label. Anchor it to 0:32, where the video says: “4 support to their library. React Native Executor Torch is an open-source project from a company called Software Mansion, and it lets you run machine learning models directly on the phone using PyTorch's Executor Torch runtime. The repo...”
02
Source
Use "Source" to locate the part of the creative automation workflow the video is demonstrating. Ask what changes in your real setup if this claim is true. Anchor it to 3:02, where the video says: “natively, which is the part that lets the model actually do things instead of only chatting, and they hold a lot of context, up to 256K tokens on the larger sizes. They also support over 140 languages, which...”
03
Generation
Turn "Generation" into the reusable artifact for this lesson: A creative workflow board with critique criteria and review checkpoints. This is where watching becomes something you can inspect and reuse.
04
Selection
Use "Selection" as the application surface. Decide whether the idea touches a browser flow, a local file, a model choice, a source document, a UI, or a review step.
05
Edit
Use "Edit" to prove the lesson. The evidence should connect back to the video title, transcript anchors, and a concrete output, not a generic best-practice claim.
06
Taste Review
Use "Taste Review" to carry the idea forward: save the prompt, checklist, diagram, or operating rule that would make the next agent run better.
Example
Source-backed work packet
Convert the video into a scoped task that includes the transcript claim, target workflow, acceptance criteria, and proof. The output should be a creative workflow board with critique criteria and review checkpoints..
Example
Claim vs. demo brief
Separate what the speaker claims, what the demo actually proves, and what still needs outside verification before you adopt the workflow.
Example
Teach-back module
Transform the lesson into a definition, a mechanism diagram, one misconception, one practice exercise, and a check-for-understanding question.
Do not learn it wrong
Treating the title as the lesson without checking what the transcript actually says.
Letting the prompt drift into generic advice that could apply to any video in the playlist.
Copying the tool setup without identifying the operating principle that transfers to your own stack.
Skipping the artifact, which means the learning never becomes operational or inspectable.
Do not count this as learned until these are true.
01
State the transcript-backed claim in your own words: A technical walkthrough of running Google's Gemma 4 edge model fully offline inside a React Native app via the ExecuTorch library, covering the memory math behind its quantized formats and the useLLM hook that handles loading, streaming, and on-device tool calling.
02
Explain the practical stakes without hype: New playlist item from AI Stack Engineer; queued for transcript-backed review, topic mapping, and a practical learning artifact.
03
Map the idea onto the Brief -> Source -> Generation -> Selection -> Edit -> Taste Review sequence and name the weakest link.
04
Produce the artifact and include the evidence that proves it: A creative workflow board with critique criteria and review checkpoints.
Put it into practice
Give this grounded prompt to Codex or Claude after watching.
You are helping me turn one specific YouTube video into real, durable learning.
Source video:
- Title: Gemma 4 + React Native: Run AI Fully Offline In Any App No Cloud, No API
- URL: https://www.youtube.com/watch?v=LuSfs__jldQ
- Topic: Creative Automation
- My current learning frame: Build a small React Native screen that loads the E2B Gemma 4 model with vision enabled, feeds it a photo of a real flyer or receipt, and wires a single tool-calling function so it can act on what it reads, entirely offline.
- Why this matters: New playlist item from AI Stack Engineer; queued for transcript-backed review, topic mapping, and a practical learning artifact.
Transcript anchors from this exact video:
- 0:32 / Evidence 1: "4 support to their library. React Native Executor Torch is an open-source project from a company called Software Mansion, and it lets you run machine learning models directly on the phone using PyTorch's Executor Torch runtime. The repo..."
- 3:02 / Evidence 2: "natively, which is the part that lets the model actually do things instead of only chatting, and they hold a lot of context, up to 256K tokens on the larger sizes. They also support over 140 languages, which..."
- 5:04 / Evidence 3: "actually drop this into an app because it's simpler than you'd expect. React Native is the framework a lot of teams use to build one app that runs on both iPhone and Android from a single code base."
- 6:36 / Evidence 4: "an on-device model useful instead of just a chatbot. You define a tool, basically a function with a name, a description, and its parameters. You hand that definition to the model. When the model decides it needs that..."
- 8:12 / Evidence 5: "audio. So, that's your starting point if you want to build with this today. All right, so that's it from the video, and I hope you enjoyed it. If you did, please like this video and subscribe to..."
Your task:
1. Use the transcript anchors above as the primary source packet. If you add outside context, label it clearly as outside context and keep it secondary.
2. Create a source-check table with columns: timestamp, claim, what the demo proves, confidence, and what still needs verification.
3. Extract the actual teachable claims from the video. Do not invent claims that are not supported by the title, lesson frame, or transcript anchors.
4. Build a reusable learning artifact: A creative workflow board with critique criteria and review checkpoints.
5. Include:
- a plain-English definition of the core idea
- a diagram or structured model using this sequence: Brief -> Source -> Generation -> Selection -> Edit -> Taste Review
- 3 concrete examples that apply the video idea to real agentic work
- 2 failure modes the video helps prevent
- a checklist I can use the next time I run Codex or Claude
- one practical exercise with a clear done signal
6. Add a "learning transfer" section: what changes in my workflow tomorrow if I actually learned this?
7. Add a "source check" section that cites which transcript anchor supports each major takeaway.
Quality bar:
- Make this specific to "Gemma 4 + React Native: Run AI Fully Offline In Any App No Cloud, No API", not a generic Creative Automation essay.
- Prefer operational examples, failure modes, and reusable artifacts over broad definitions.
- Call out uncertainty instead of smoothing over weak evidence.
- If evidence is weak, say what transcript segment or timestamp needs review instead of guessing.
- Finish with a concise artifact I could paste into my learning app.
Misconceptions
What to stop believing.
Creative AI removes the need for taste.
It increases the need for taste because output volume explodes.
The best prompt is enough.
References, critique, iteration, and post-production matter just as much.
Practice studio
Learning only counts when you make something.
01
Transcript evidence map
Separate what the video actually says from what you already believe about the topic.
3 source-backed takeaways with timestamps, confidence, and a transfer note.02
One useful artifact
Apply the video to a real workflow and produce a creative workflow board with critique criteria and review checkpoints..
A reusable artifact with a done signal and one verification step.03
Teach-back card
Explain the lesson to someone who has not watched the video yet.
A 90-second explanation, one diagram, one example, and one misconception to avoid.
Recall check
Answer first, then reveal — without rewatching.
Why does the video call Gemma 4's Apache 2.0 license 'a bigger deal than it sounds'?
How did Google get the E2B model's footprint down to about 1GB, and what happens if you only need text?
What are the two hardware acceleration delegates the ExecuTorch React Native library uses, and what do they enable?
Source shelf
Use the video as a doorway, then verify with primary sources.
