@gokulr @chamath Compared with https://t.co/11O2TcyUsp?
— Abhay 🇸🇬🇮🇳 (@Abhay08)
Apr 6, 2026
Category: Twitter Page 4 of 84
@chamath I use qmd + obsidian with a cron to update my knowledge base of all convs. Whenever I start a new conv, have found the agent/chat references my KB as I have talked my chats/agents into it. I use Claude, CC and openclaw only though and it works as I expect.
— Abhay 🇸🇬🇮🇳 (@Abhay08)
Apr 5, 2026
RT @ClementDelangue: llama-server -hf ggml-org/gemma-4-26b-a4b-it-GGUF:Q4_K_M
openclaw onboard –non-interactive \
–auth-choice custom-…— Abhay 🇸🇬🇮🇳 (@Abhay08)
Apr 5, 2026
@pbteja1998 just use qwen3.5-27B-GGUF locally as a primary model and routing rules on top if you are doing anything fancy…works like a charm and your wallet will thank you
— Abhay 🇸🇬🇮🇳 (@Abhay08)
Apr 5, 2026
Guess Claude code proved this way earlier when it was more performant with sonnet 4.6 than other frontiers, it’s the harness fgs!! https://t.co/3dl7NMODr7
— Abhay 🇸🇬🇮🇳 (@Abhay08)
Apr 5, 2026
Not impacted with anthropic cutting subs from claws. If you have been using your agents regularly, using anthropic 100% was never a sustainable option as it was clearly $200++ bill a month, so I have been using locally hosted Qwen3.5 with other options. No complaints. Kimi2.5 has https://t.co/53quTIemdT
— Abhay 🇸🇬🇮🇳 (@Abhay08)
Apr 4, 2026
OSS models are closing the gap, and with GGUF, NVFP and MXL, local inference is gaining wider adoption. Frontier models will still be the apple of the industry https://t.co/uH662icIlC
— Abhay 🇸🇬🇮🇳 (@Abhay08)
Apr 3, 2026