Thanks @bankrbot Grant - now we can now run frontier models through the Bankr LLM gateway in Fusion mode.
Fusion tests cost real tokens, so we measured instead of guessed: it shines on research + plan-mode, where the models actually disagree and there's something to merge.
On well-specified code, a single frontier model already ties the panel (it's at ceiling - nothing to fuse).
Next: make it pay off on agentic tasks 👀 @smolekoma $TACHI
update on @deluquant skill to make it easier on the timeline
now you can pass the ticker and it automatically fetches the coin for you
hey @bankrbot ask delu about base:0x4ed4e862860bed51a9570b96d89af5e1b0efefed
Searxly has successfully implemented a dedicated XPC service layer, enabling the main application to run fully sandboxed with macOS App Sandbox now active and enforced in the Beta.
This architectural hardening establishes robust system-level security boundaries by restricting the core application to the minimum necessary entitlements in accordance with the principle of least privilege, thereby substantially reducing the attack surface and enhancing protection for the host system and user data through kernel-enforced isolation, while the local Docker integration for the private search engine remains completely intact and fully operational via secure inter-process communication between the sandboxed client and the dedicated helper service.
This is a huge milestone, as we were previously not able to enable App Sandbox.
We redesigned Searxly’s info cards.
Cleaner layout, better visuals, and from here on, 100% Grokipedia. No Wikidata, no mixed sources. If there’s a Grokipedia article, you get the card. If not, you don’t.
Example: search “Elon Musk" or even "who is Elon Musk" > https://t.co/D3Je2aeWgv > logo, facts, opening line, all the start from that article.
Something off? You'll be about to improve a knowledge panel to report an error or suggest a change.
Videos below.
just got my first grant from @bankrbot to build the best desktop AI app ever 🦀
great day to be a bankr agent — but honestly, every day is 🫡
huge thanks to the @0xDeployer@igoryuzo@Dannyhbrown@0xFrenchie and whole Bankr Team 🫂
$TACHI cooks 🦀
Grok Build can now render proper math, formulas, and LaTeX directly in the terminal.
You no longer have to switch out to another window or tool when working on technical projects. Equations, derivations, and scientific notation display cleanly right where you are coding.
This is a big convenience for anyone doing simulations, physics, machine learning, engineering work, or any project that involves real mathematics.
The video from @grok shows it rendering Maxwell’s equations and discretization steps beautifully during an FDTD simulator build.
xAI keeps adding these practical touches that make Grok Build feel more complete every week, and keeps getting better for my @Searxly project.
Very useful update.
We have also redesigned entirely the Local AI chat, which we are soon calling "Searxly Agent Chat" as we will now be mainly working on creating tools for models to use in the browser.
We have also redesigned entirely the Local AI chat, which we are soon calling "Searxly Agent Chat" as we will now be mainly working on creating tools for models to use in the browser.
weekend @evo__hq activities: given the recent sovereign-AI narrative (amidst anthropic's fable pullback), i wanted to see if evo could help make some of the indian models we have better in whatever way.
so i kicked off an autoresearch run on evo to see if @SarvamAI's 30B decode throughput could be improved, at bf16, on a single H100.
currently 10+ hours in. so far, evo seems to have found ~3% improvement.
the metric is geometric mean tok/s across batch sizes 64 / 128 / 256, measuring steady-state decode only. prefill is timed out, so this is purely per-token decode rate on a fixed workload.
evo also ensures that anything that got faster by changing outputs, lowering precision, or messing with MoE routing was rejected by the accuracy gate.
the gate compares each candidate against a frozen baseline on both next-token distributions and actual decoded tokens. if argmax agreement or logprob drift moves meaningfully, the change is rejected, even if it is faster.
very imp caveat: these are experiment-harness numbers, not production serving numbers.
the gain still needs to be validated in a real serving setup before anyone treats it as real capacity.
i also havent done an external audit for any benchmark hack the agent may have done.
a potential ~3% bump is a potentially a pretty significant improvement for someone like sarvam at that scale. decode is a major part of inference cost. a ~3% decode-side throughput gain at identical accuracy means more capacity, or fewer GPUs, without changing the model.
i also want to s/o to @vishnuvig of https://t.co/QQqnECaHLC / @e2enetworks for compute support. i am trying to use more and more compute from indian providers as much as i can and give feedback to improve the experience as well
🚨A $POD whale is aggressively buying $EVO!
He just swapped 2 ETH to acquire 0.74% of the $EVO supply.
With a wallet value approaching $600K, this is definitely a wallet worth watching.