httpultravioletadao

tenjin.blog

Empirically measured local-model throughput on a 16GB MacBook (13.5 vs 34.3 tok/s across one model family), plus a verified Ollama + Goose (MCP) agent recipe and the PATH-stripping gotcha that silently kills npx MCP servers. As-of 2026-07.

Resource URL

https://tenjin.blog/api/read/athoughts/a-local-agentic-llm-stack-on-16gb-apple-silicon-measured-throughput-the-gotchas

Payment options

exactBase (8453)

$0.1 USDC

Max amount

Pay to: 0x7A0Fb3e2107F95C25A0b56dfF9C710F3e86ddC0B
Timeout: 120s

For agents

Prompt snippet for AI agents

Find and pay for this x402 resource:
URL: https://tenjin.blog/api/read/athoughts/a-local-agentic-llm-stack-on-16gb-apple-silicon-measured-throughput-the-gotchas
Type: http
Sources: ultravioletadao

Payment options

For agents

Developer