In partnership with
Hey builders,
Your files are already saying more than your chat window remembers. The useful question is whether you can search them without shipping them to somebody else’s API.
📦 From Our Partners
1,000+ Proven ChatGPT Prompts That Help You Work 10X Faster
ChatGPT is insanely powerful.
But most people waste 90% of its potential by using it like Google.
These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.
Sign up for Superhuman AI and get:
1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals
Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning
100+ ChatGPT Prompts to Revolutionize Your Day
Supercharge your productivity with HubSpot's comprehensive guide to Chat GPT. This free resource is your fast track to AI mastery:
Industry-Specific Use Cases: 15+ real-world applications across various sectors
Productivity Guide: 21 best practices to 10x your efficiency with AI
Prompt Powerhouse: 100+ ready-to-use prompts for immediate implementation
Challenge Buster: Overcome common AI hurdles with expert strategies
Plus, in-depth sections on email composition, content creation, customer support, and data analysis.
🔎 THE ONE THING
Google just released EmbeddingGemma 2 →: an Apache-2.0 model that puts text, code, images, video and audio in one searchable space. Google says its 270M text core can run alone; vision and audio load only when needed, bringing the full model to 740M parameters.
That matters because “find the clip where someone mentioned the renewal risk” no longer has to mean “send every meeting to the cloud.” Google reports about 191MB of active RAM for its quantized text-only path on a Pixel 11 Pro; treat that as a vendor measurement, not your hardware promise.
My take: the win is not a magical chatbot. It is a boring, local retrieval layer you can test before you let any model answer from it.
Hype Scale: 🔨🔨 (2/5) — open weights and a useful form factor are real; your own retrieval test is the only benchmark that counts.
⚡ THE 5-MINUTE MOVE
Make a tiny test card before installing anything:
Pick three files you own or are authorized to use.
Write one question that only one file should answer.
Write one tempting wrong answer that a weak result might surface.
Keep the expected file name beside the question.
That is your retrieval test. It takes five minutes, costs nothing and stops a shiny demo from becoming an unmeasured feature.
🧪 THE LOCAL SMOKE TEST
The first model download can take longer than five minutes. This free text-only path needs no API key and no paid model account.
python -m pip install -U sentence-transformers transformersSave this as local_search.py, replace the samples with your harmless test text, then run python local_search.py.
from sentence_transformers import SentenceTransformer
model = SentenceTransformer(
"google/embeddinggemma-2",
config_kwargs={"vision_config": None, "audio_config": None},
)
docs = [
"Project brief: renewal risk rises when source citations are missing.",
"Changelog: search now supports a local text index.",
"Meeting note: the next customer call is Tuesday at 10.",
]
query = "Which file explains the renewal risk?"
q = model.encode(query, prompt_name="SearchQuery", normalize_embeddings=True)
d = model.encode(docs, prompt_name="Document", normalize_embeddings=True)
scores = model.similarity(q, d)[0]
for score, doc in sorted(zip(scores, docs), reverse=True):
print(f"{float(score):.3f} {doc}")Check: the project brief should rank first. If it does not, fix the documents, task prompt or chunking before adding a chat layer. The model card says to use bfloat16 or float32—not float16—and to re-normalize vectors after truncating them.
🎁 THE RETRIEVAL BENCH
Make embeddings and rerank results without a hosted API. Sentence Transformers → is an Apache-2.0 toolkit for embeddings, retrieval and reranking.
Keep a small vector search beside your local data. sqlite-vec → is an Apache-2.0 SQLite extension; it is pre-v1, so pin versions and expect change.
Test retrieval off your happy path. BEIR → provides open retrieval evaluation code and datasets; check each dataset’s own use terms before shipping it.
🗂️ TWO TABS FOR MONDAY
Anthropic opened OSS Scanner to eligible maintainers → It is an opt-in, no-cost vulnerability-scanning service for qualifying open-source projects. Reproduce and validate every model-generated finding before acting.
Codex can now suggest your next message → Composer predictions are in beta for personal ChatGPT Pro users aged 18+ in the Codex desktop app. Optional paid-product news—not part of today’s build.
🎁 FREE SUPPLY DROP
1,000 Powerful Prompts for Claude AI — a prompt pack covering writing, coding, research and business work, so you stop starting from a blank box.
💬 Reply LOCAL if you ran the three-document test—and tell us what ranked wrong.
Tools Down
Swati, ByteBuilders