@OmarShahine@openclaw Hope Microsoft is big donors to @Openclaw foundation. A non profit OSS turnt to a for profit for every 365 client. And yes please say this more instead of saying autopilot.
@BenjaminBadejo Too emotionally attached on the vendor man, remember they all a means to your goal. You dont know what you missing with opus and 2 months is a century of things you could build better, because of opus.
Running already oMLX and glm5.3 flash on an @openclaw gateway at an M5 ultra 256gb RAM , now I can wire Tensorfold as well. Openclaw native harness helps a lot where memroy native tooling retrieves the sqlite per agent and per gateway plus wiki memory native tools. Super powers for local models
Well @AnthropicAI is having issues today, so we quickly switched providers in our @openclaw enterprise deployment and... continued on like (almost) nothing happened.
That's the power you have when you use OSS tools and not get vendored locked in.
@verrsane Why you disregard openclaw? It literally has the better native harness to be agnostic on llm providers while its configs and knobs is a joy to create the most specialized business workflows.
If it all Hermes is way inferior for business
@ceckert@vincent_koc@openclaw How you find the openclaw native harness? Do you agree mastering this and tapping any LLM provider on top is where you get the max of your claw along with of course being a master of your config knobs?
Four OpenClaw developers made @linuxfoundation’s Top 100 contributors list:
#1 @steipete
#7 @vincent_koc
#33 @obviyus
#57 @dallinfinite
569,649 contributions in a year. 13.8% of the Top 100 total.
Thank you, open-source community.
https://t.co/hk8IFZAHpI
@NousResearch@steipete@pubity The inferiority complex runs wild accross all nous research and its paid army of nothing burger users. Openclaw is being used from people really knowing what own your harness means. Its so tiring a for profit org using blunder to the only real OSS for agents.
@dandolalata@steipete Normally there is a button under the attachement to click and converts to text . Seen it in the Mac OS app and felt same how codex does and you just click and expands as text
Deepseek 4.1 is blazing fast with M5 Ultra 256GB. Preliminary result from MLX-Serve, @ddalcu got to 2,000+ tok/s prefill and 80+ tok/s gen ...and is still climbing!
Next release will be wild.
@fagamericano@openclaw@AnthropicAI@OpenAI Fully get your frustration, up to speed now with versions? Since I migrated, all upgrades now are very smooth. Canary soak used to take "ages", not any more. Only people maintaining pre v9-v10 era can get the feeling instantly.
Yes at this stage, cost of this set up does not satisfy fully local set up except if you are an enterprise or with super sensitive info (patients data etc) other than that, knowing what runs with inference to frontier models while you opt out from any training on your prompts and what runs 24/7 as continuous monitoring via local inference is where you nail the split.