Non-coder builds layered memory architecture: 20k tokens tracks a year of agent conversations
matteoianni · reddit · 2026-09-11
A developer with zero software background shares an open-source hierarchical memory architecture for long-running LLM sessions, now a Codex CLI plugin and standalone Briglia CLI assistant.
Core ideas
- Skip memory plugins: agents can't reliably know when to look things up, so memory must live in context; the relevance decision can't be delegated to a weaker model either.
- Messages are the ideal memory distillation — a day (10k tokens) compacts into 1k dense summary tokens with little loss.
Layering
- Keep 1k/day for recent days, merge every 4–5 older days into another 1k summary with chronological IDs for retrieving full text.
- 10k tokens covers a month; 20k tokens tracks a year of interactions.
Constraint: all work must run on one continuous main thread (subagents may parallelize) so the message chronology stays coherent. Both projects are on GitHub (permaevidence/codex-plugins, permaevidence/briglia-cli).
More from coding & agent
- Alex Townsend posts 200 open problems in numerical linear algebra for humans and AI agents — IgorCarron · 2026-09-11
- Kimi K2.8 Preview rolls out: near-K3 coding performance, 1M context for all tiers — teortaxesTex · 2026-09-11
- Looking for a classifier of software engineering task shapes to pick models per task — StewartalsopIII · 2026-09-11
- Steal this idea: prompt-to-hardware where agents assemble custom devices — paraschopra · 2026-09-11
- Model Is the Least Interesting Part: A Guide to Six Core AI Architectures from RAG to Multi-Agent — goyalshaliniuk · 2026-09-11
- Warp's six non-engineering teams all run on Linear and Claude Code — mon__lim · 2026-09-11