Local Agentic Memory Kernel Released
Brain-AI-Mapping · reddit · 2026-07-15
The author released Brain-AI Memory v0.4.0, an MIT-licensed, installable, local, and provider-neutral reference kernel designed to help long-running agents better manage memory and failure boundaries.
Segregated Memory Layers
- episodic events
- semantic knowledge
- procedural rules
- exact state
- entities and relations
- explicit lifecycle decisions and checkpoints
Form Factor & Capabilities
- Provides a Python API, CLI, and MCP server
- Features an optional downstream bridge to inspect proposed actions and run host-provided fallback sequences
Author's Emphasized Boundaries
- This is neither a new retriever nor a biological simulation
- RAG, rules, hooks, databases, and harnesses are existing primitives
- The core contribution is testing whether explicit ownership, failure boundaries, entity scope, and lifecycle contracts make failures in long-running agents easier to diagnose and control
Current Limitations (Not Automated)
- Does not auto-import Claude Code / Codex transcripts
- Does not auto-assemble token-budgeted model context
- Does not schedule lifecycle operations
- Does not enforce host-wide actions
- Does not physically delete retained source data; these remain the host's responsibility
Negative Results
On 500 cleaned LongMemEval-S questions, a 96-keyword pointer reduced indexed source text by 93% compared to full-session BM25, but answer-session recall@3 dropped from 86.1% to 71.0%. The author emphasizes this is strictly a retrieval-only result and does not indicate answer quality or being "better than RAG."
Additionally, a deterministic ablation checked 10 out of 20 contract mechanisms: the full condition passed 20/20, while the flat control passed only 1/20. However, this is merely evidence of conformance, not proof of better LLM answers.
More from coding & agent
- Coding agents are heading toward an AI-writes, AI-reviews, human-approves workflow — aftahi_ai · 2026-07-22
- oMLX 0.5.2 adds Mac menu-bar stats, low-bit decode kernels, and faster downloads — awnihannun · 2026-07-22
- GitHub review bot hits its PR limit and forces a 39-minute cooldown — DanielLockyer · 2026-07-22
- Max reasoning effort appears to be mobile-only in Codex Remote, not desktop — GabGarrett · 2026-07-22
- A Reddit demo argues online stores should expose carts and pricing through MCP — gelembjuk · 2026-07-22
- Open-source AI SDK provider routes Vercel apps through a local Codex subscription — lgrammel · 2026-07-22