Running 3 agents + 10 subagents in 16GB RAM with 5GB to spare
Teknium · x · 2026-09-26
Developer RecursiveIntell shows 3 Hermes agents plus 10 subagents and 3 Chrome tabs running on a 16GB (14 usable) machine with 5.25GB RAM still free — something he assumed needed 64GB. He traces the journey from "how much can I run in 16-32GB" to building deterministic model routing, semantic memory injection, governed tool execution, receipts/evidence, context management, heterogeneous local compute (merging his crate with NVIDIA PAIR, where he found and patched a bug merged into the NVIDIA/PAIR repo), and resumable agent execution. The goal: agent systems that are more efficient, inspectable, and local-first. Teknium amplified the thread.
More from coding & agent
- Real-world case: Opus 5.5 clearly beats GPT-6 Sol on architecture-level coding — DataLearnerAI · 2026-09-26
- Claude keeps killing its own grep and shell processes, and prompts don't fix it — SebastianNehrd2 · 2026-09-26
- Dev builds NVIDIA Blackwell GPU 3D animation in one HTML file with Claude Opus 5.5 — EricBuess · 2026-09-26
- Tech conference wrap-up: AI coding is evolving into the 'agentic software factory' — ahahabbak · 2026-09-26
- Xiaomi MiMo open-sources its RL environments and training code on a fork of verl — eliebakouch · 2026-09-26
- We already rely on AI to police rogue agent behavior, and that's a worrying sign — JeffLadish · 2026-09-26