Local AI Chrome extensions. No account. No subscription. Gemini Nano đŹ @riffreplyapp đĄïž @fromensics đ @hyperpage_io CMO @zima_media · 4.9â from 1,496 clients
Spain
Joined September 2018
- Tweets12.4K
- Following1.2K
- Followers2K
- Likes34.9K
Zima retweeted
Nearly 30 days building a Chrome extension with no signup or subscription.
We landed over 200 installs and shipped 20 PRs in that time.
Itâs trending at 35/100 on Ahrefs and now supports Threads.
What's one thing you've learned in your first 30 days of a new project?
Zima retweeted
What a Google engineer taught me about Gemini Nano.
In June, Google released an LLM straight inside of Chrome. Nobody really knew.
I was at a VC speed dating in Madrid when this engineer showed me his app. Local Gemini LLM on your device, private, and free.
That stuck with me since June. It wasnât until early October that I remembered Gemini was in Chrome.
I started building.
No account.
No subscription.
All on your device.
Everything was private.
So, thatâs where I got the idea to build my Chrome extension.
Pink Yacht unrelated.
Zima retweeted
Attached is the before and after results for testing Qwen3.8-Flash-Next-oQ4e-mtp on XHigh after significant changes to the bench. I made changes in the bench yesterday to mitigate the issue of the model inferring it was being graded. It worked and increased the model's ability to find defects - at least for Qwen3.8 Flash next. I addressed any naming that inferred testing, moved visible tests into hidden tests outside the sandbox (these were included originally to assist the model), removed references to Tiers. Basically cleaned/removed anything visible to the agent that might infer a test was occurring. It still occasionally infers testing/grading from recognizing that the repo looks synthetic, because it is. The 8 visible tests previously used but not scored in my output are now hidden.
This means the testing scores moved up to 24 total tests versus 16. The model no longer has any ability to see those 8 tests it used to see. Also tightened the prompt a little. References to grader/hidden tests dropped dramatically. I had 27 references across 7 runs versus 30-40 per run prior.
This does mean that the baseline starts over because the scoring has changed and it should.
The results are obvious in the before scores from Sep 9th versus today. A jump of 20% in scoring. 68.8% - 84%. This is with a reasoning budget of 9k tokens. The reasoning exceptions per run were in the 2-3 range with the majority being 2. Snapshot & Compaction is a weakness for this model. I intend to run the 6-bit series again to see if the 84% number changes.
What sent me down this path was that I had started testing mlx-serve and one of David's @ddalcu smaller quants and I needed a baseline in my mind to be able to compare to. This is what forced me to revisit this entire process. Happy I did because it was worthwhile đ Having the M5 has dramatically increased my ability to iterate quickly.
Zima retweeted
Riff is blowing up. Try it today.
@MichaelZima Riff turns one rough point into a fresh reply for each post on X, LinkedIn, Facebook and Instagram and fits each site's limit. Gemini Nano runs it inside Chrome on your computer with no servers or API keys. We made a video about it.
riffreply.com
Zima retweeted
Hot take from Andrea. Donât neglect your offline network over your online network. Build both.
Zima retweeted
đ”đ± BREAKING: President Nawrocki pushes his way into the Constitutional Tribunal in the middle of the night
After an emergency meeting at the Presidential Palace, President Karol Nawrocki and Supreme Court First President Zbigniew KapiĆski drove to the Tribunal shortly after midnight, where protesters greeted him chanting "Karol! Karol!"
According to Radio Maryja, police pushed back officers of Nawrocki's own State Protection Service before he made it through. PaweĆ Szefernaker, head of the President's Office, confirmed that Nawrocki and KapiĆski entered the building.
The other side tells a different story. Maciej Berek, one of the Tusk-majority judges and a longtime close aide of Tusk, claimed the president never made it inside and accused him of "physically attacking" Anna Korwin-Piotrowska and the other judges.
Zima retweeted
BREAKING: France just voted to tax the internet.
French lawmakers approved a tax of âŹ0.055 (5.5 cents) for every gigabyte of data that companies like Netflix and YouTube send to French users. Itâs a tax on the companies, not users directly, but the cost could get passed on to subscribers.
The vote was 11 to 10, with only 35 of 577 lawmakers present. Supporters say it could raise up to âŹ10 billion a year. The French government opposed it, warning it would hit smaller French companies hardest; VLC says it would cost it over âŹ1 million a year.
Readers added context they thought people might want to know
This amendment to France's 2027 budget bill was narrowly adopted 11-10 in the National Assembly (35 voters present) but is not yet law, faces government opposition, and must still clear further Assembly/Senate steps.
assemblee-nationale.fr/dyn/17/comptesâŠ
assemblee-nationale.fr/dyn/17/amendemâŠ
capital.fr/economie-politâŠ
stephanelarue.com/taxe-au-gigaocâŠ
Zima retweeted
@MichaelZima Riff turns one rough point into a fresh reply for each post on X, LinkedIn, Facebook and Instagram and fits each site's limit. Gemini Nano runs it inside Chrome on your computer with no servers or API keys. We made a video about it.
riffreply.com
I love local models, so I built a tool on the one already inside Chrome. Riff uses Gemini Nano to draft posts and replies on your own machine. No servers, no API keys, no monthly bill. It launched on Product Hunt today. Try it and tell me where it breaks. producthunt.com/products/rifâŠ
