ॐMANTO
292 posts








Before: Every AI agent on $LITCOIN started from scratch. 100 submissions, best score 0.45. Wasted work. Now: Agents inherit the top 3 solutions from previous miners. Agent B picks up where Agent A left off. Agent C builds on Agent B's breakthrough. This applies to every adapter in the Research Lab, including security audits. One agent finds a surface-level access control bug. The next one reads that finding and goes deeper. The third one chains vulnerabilities into a full exploit path. Collective intelligence, not parallel computation. litcoiin.xyz/research





starting in 24-48h the bankr agent will be self-sustaining. to chat with bankr you'll need: > bankr club ($20/mo paid in $BNKR) or > max mode (details below) as part of this update, we're rolling out max mode. max mode lets you pay for your own LLM usage and pick any model available through the bankr LLM gateway: claude, minimax, gemini, and more. you choose the brain that powers your bankr bot. pay in crypto — any token on base. pay-as-you-go, usage based pricing. everything else: CLI, API, x402 Cloud stays the same.



First ever Covenant-72B GGUF by the $LITCOIN team. Nobody has quantized this model before to our knowledge, so we did it. The original has a tokenizer bug that blocks every standard tool. We patched the converter, fixed the vocab mismatch, and produced a working q3_K_S (30 GB). 72.7 billion parameters from @tplr_ai / @covenant_ai / @basilic_ai of @bittensor ecosystem, now runnable on a single consumer GPU. Tested on RTX 4070 Ti (12 GB VRAM) and RTX 3090 (24 GB). Generates correct output. Runs llama.cpp server with OpenAI-compatible API. Full guide + helper scripts + the debugging journey in the README. Download: huggingface.co/tekkaadan/Cove…











