EVO

this is what we at @evo__hq do and believe strongly in :)

excited to see what you all do with evo !

@Andy89554819743 💯 articulation of what EVO does !

EVO Series - The New Era of Self-Improving AI Part 2: How EVO Actually Works The idea behind @EVO__HQ sounds complicated at first, but the core mechanism is surprisingly simple. Imagine you have an AI agent solving tasks. Normally, the agent follows one fixed set of instructions. $EVO changes that by turning the process into continuous experimentation. Here’s the simplified version: EVO creates multiple variations of the agent’s workflow or “Skill” Each version approaches the task slightly differently All versions are tested in parallel on the same benchmark The system measures which one performs better The strongest approaches survive and continue evolving Weak ones get discarded Then the loop repeats again and again. What’s important is that the underlying AI model does not change. The improvement comes from optimizing: the workflow the reasoning process the task strategy the structure around the model In other words: EVO is trying to improve how AI thinks and operates, not just how powerful the model is. That’s what makes this direction interesting.

this close to 500 followers !

@gumterver100 @bankrbot @base @baseapp cool !

Top @bankrbot launches last week We saw some of the biggest launches on Bankr last week. Some of them already made it into the top 10 coins on Bankr. The trenchers are bringing more builders, supporting them, and it is all happening on @base.

we have been busy shipping 🚀🚀 now accepting early users to our hosted platform. sign up here - evo-hq.com

bullish. AI's 4 minute mile is here. model capability x good harness + tight verification loops is where its at i am so excited for what's possible down the line, esp with autoresearch. one of the reasons i am all-in on @EVO__HQ

related : one of the biggest reasons I’m bullish on @evo__hq / autoresearch is how often people are amazed the first time they run it. they’ll DM or email me saying they found huge speedups or improvements almost immediately. wild part is that you don’t even need to push the frontier of research for this to be valuable. there is still so much software running with suboptimal implementations, a lot of the low-hanging fruit could be found and fixed today by simply running one autoresearch loop inside the system.

looked at the latest npm and pypi stats today. as of this evening, evo has been used to optimize atleast 10,000+ individual projects / repos. all of this without a single paid ad. thank you to everyone who has tried evo, shared feedback, reported bugs, or just told me what felt confusing. if you have feedback, my DMs are open.

For all you terminal nerds, a quick sneak peek at a QoL improvement shipping in the next version of @evo__hq You can now track your autoresearch runs right inside your terminal, no need to open the dashboard (which evo already ships with). Also in the works: better steering of long-horizon autoresearch runs, landing in an upcoming release. What else would you like to see in evo ?

awww, first fishing attempt 🫶🫶

i am pleasantly surprised talking with a lot of @evo__hq users who have found great success applying it to frontier/novel research problems. the interesting thing is evo isn't even tuned for those sort of tasks right now !! my original objective was just to make autoresearch much easier for a wider audience to apply on their own systems or ML setups. and it still seems to be impressing a lot of folks applying it in serious scientific research. lots of alpha yet to capture.

It’s been an interesting couple of days. Most of you know I’ve been building evo in public over the last few weeks. The goal is simple: make evo the go-to platform for autoresearch and optimization. Whether it’s code, models, agents, or internal systems, evo should make it easier and cheaper for organizations to keep improving them automatically. Since launch, evo has now been installed and used on nearly 6,000 systems and has crossed 700+ stars on GitHub. The user conversations have honestly been overwhelming. People are reporting scientific SOTA results, optimizations they did not expect, and improvements in systems they had already spent a lot of time tuning manually. You can check out the project here: github.com/evo-hq/evo While all this was happening, someone anonymously minted $evo on @bankrbot and assigned me as the creator. As someone who is not crypto-native, I had to spend some time understanding what this means. From what I understand, I receive creator fees, which are a small percentage of the volume traded. I’ve been told this is standard and above board. So with that said: I’m all in on $evo It will be the formal token associated with the open-source project I’ve been building. I’m plan on holding my allocation and not selling, regardless of price. I’ll also be claiming the creator fees on Bankr( once it is back online! ). Every bit of that will go back into evo: product development, compute costs, keeping the project running, and giving back to the power users who are actually using it. I also want to use part of the fee proceeds to support charities chosen by a small cohort of evo users who are willing to get on a short call with me every month and help shape where the project goes. More on this soon. The better evo gets, the bigger this gets. Still early. Still building in the open. Just getting started.

Where @EVO__HQ is heading : I've been planning some solid next steps for EVO. I'll keep pushing autoresearch applications, but a lot of recent user conversations have convinced me the core optimization problem runs much deeper than I first thought. Customers don't want a one-time autoresearch run. They want their systems to stay continuously tuned. So it's time for evo to serve any optimization need an org might have: systems, code, agents, and even models. The long-term goal is for evo to become the platform teams choose to run agents 24/7 and constantly tune everything they're building.

with the latest version of evo , you can now run autoresearch loops super easily inside @cursor_ai with just two commands - /discover and /optimize that evo ships get evo on cursor with two simple installation steps $ uv tool install evo-hq-cli $ evo install cursor

one of the things i have been very curious about is how to measure and improve novelty for autoresearch run. So introducing : @EVO__HQ 's autoresearch-novelty-bench : A benchmark for measuring whether AI agents during autoresearch runs propose genuinely novel ideas built on top @PrimeIntellect's autonomous-speedrunning archive: claude Code & Codex racing on the modded-nanogpt speedrun. contains over 10k+ training runs, 600+ idea writeups, an intense 2-week burst of parallel autoresearch - we know what each agent tried, when, and whether it worked. thread

for those of you who are autoresearch pilled , or have been meaning to get into autoresearch but dont know how - I shipped evo today - a opensource Claude Code plugin that optimizes code through experiments you hand it a codebase. it finds a benchmark, runs the baseline, then fires off parallel agents to try to beat it. kept if better, discarded if worse. inspired by @karpathy's autoresearch, but with structure on top: - tree search over greedy hill-climb — multiple forks from any committed node - N parallel agents in git worktrees - shared failure traces so agents don't repeat each other's mistakes - regression gates

Been heads down for weeks - talking to users, shipping their feedback. Pleased to announce evo v0.4 is now generally available !! evo is an open-source autoresearch orchestrator: point it at your codebase and it becomes a self-improving loop - agents run experiments parallely, keep what beats the benchmark, drop what doesn't. everything new in v0.4 👇 [thread]