A personal AI agent architecture: local Llama router, whitelisted passwords, human-in-the-loop approvals
roamingandy · reddit · 2026-10-11
A Redditor laid out a detailed architecture for a personal task-handling agent and asked for feedback:
- Local Llama as router: handles simple tasks token-free and delegates to the best agent/model per task; Gitpilot in VSC manages workspaces with different profiles
- Memory & monitoring: Basic Memory records each agent's personality profile and completed tasks across sessions; Llama watches token and password usage for anomalies
- Safety design: KeePass with per-agent password whitelists; consequential actions (like sending email) require manual approval via Beeper; banking stays on a separate PC; Windows Sandbox considered for isolation
- Blast-radius control: Thunderbird limited to the last 30 emails; Digital Ocean, Mailgun and Cloudflare accounts capped with spending limits
A fairly complete blueprint for a security-conscious personal agent setup.
More from coding & agent
- Evals are replays: store tasks in R2, sandbox in Docker, and 90% of the work is measuring — danshipper · 2026-10-11
- Solo dev dilemma: self-hosted n8n or FastAPI on Cloud Run for LLM automations? — Mysterious_Profit696 · 2026-10-11
- Autoresearch agents get stuck in 'idea basins' — fork-and-flush offers a fix — menhguin · 2026-10-11
- JetBrains Air hands-on: running two parallel coding agents inside WebStorm — mhdfaran · 2026-10-11
- Vibe coding an Ori-style platformer with a giant boss chase using Opus 5.5 — chongdashu · 2026-10-11
- Dev builds tiny logger to find what eats OpenAI budget — token counts pointed to the wrong feature — Atm1n9 · 2026-10-11