Memory
SuperLocalMemory
Governed, local-first memory for agents. Every recall points back to its source, and writers whose facts keep failing verification lose influence.
CodeAGPL-3.0
Topic hub · Memory
Agent memory you can audit.
Agents forget between sessions, and when several agents share one store, a single bad writer can poison what all of them recall. Memory needs a source for every fact and a way to stop trusting writers whose facts fail.
Start here
Read it in this order.
Your progress is saved in this browser.
Reading path · 0 of 6 read
- PaperSuperLocalMemory 4.0: The Governed Memory Operating System for AI Agents
- ToolSuperLocalMemory: Governed Memory for AI Agents
- PostAI Agent Memory, Rebuilt: The SuperLocalMemory Relaunch
- PostAI Agents Don't Just Need Memory. They Need Memory Governance.
- PostHow AI Memory Actually Works
- PaperSuperLocalMemory V3.3: The Living Brain -- Biologically-Inspired Forgetting, Cognitive Quantization, and Multi-Channel Retrieval for Zero-LLM Agent Memory Systems
Tools
What to run.
Coordination
SLM Mesh
Peer-to-peer messaging between agent sessions, with file locking and shared state so parallel agents don't overwrite each other.
CodeElastic-2.0
Gateway
SLM MCP Hub
One hub process in front of every MCP server and every AI client, with federated tool discovery, a shared cache and cost telemetry.
CodeAGPL-3.0
Research
The papers.
SuperLocalMemory 4.0: The Governed Memory Operating System for AI Agents
Governed memory with scoped component evidence and disclosed negative results
SuperLocalMemory V3.3: The Living Brain -- Biologically-Inspired Forgetting, Cognitive Quantization, and Multi-Channel Retrieval for Zero-LLM Agent Memory Systems
Seven retrieval channels and zero LLM calls in the hot path; 70.4% on LoCoMo in zero-LLM mode
SuperLocalMemory V3: Information-Geometric Foundations for Zero-LLM Enterprise Agent Memory
75% LoCoMo accuracy without cloud calls, from a four-channel retrieval architecture
SuperLocalMemory: Privacy-Preserving Multi-Agent Memory with Bayesian Trust Defense Against Memory Poisoning
Shared memory is an attack surface — Bayesian trust scoring that degrades a poisoned writer instead of trusting it
Glossary
Terms you'll meet.
- Memory poisoning
When one compromised or careless agent writes false facts into a memory store that other agents share, so every agent that recalls from it is misled.
- Writer trust
A per-source trust score, modelled as a Bayesian posterior and updated by downstream outcomes, so a writer whose facts keep failing verification loses influence over retrieval.
- Zero-LLM retrieval
Memory retrieval that ranks results by computation, such as distances on a statistical manifold, instead of calling a language model in the hot path.
- Multi-channel retrieval
Searching memory through several independent channels at once and fusing their results by reranking, so no single index decides what is recalled.
- Memory decay
Deliberate forgetting with decay curves and consolidation, so old facts stop crowding out current ones as a memory store grows.
- Provenance
A record of where a remembered fact came from, kept alongside it so any recall can be traced back to its source.
Writing
Posts on memory.
What's changing
AI Agent Memory, Rebuilt: The SuperLocalMemory Relaunch
SuperLocalMemory is relaunched: one governed, local-first memory brain for Claude Code, Codex, Cursor, ChatGPT and every agent that speaks MCP or runs a command.
18 min read
Research explained
AI Agents Don't Just Need Memory. They Need Memory Governance.
The SuperLocalMemory 4.0 paper argues that durable agent memory is operational state—and proposes governed writes, inspectable recall, explicit local-first boundaries, and verifiable operations.
11 min read
How we built it
How AI Memory Actually Works
A ground-up explanation of AI memory: vectors, time, graphs, ingestion, multi-channel retrieval, controlled forgetting, isolation, and safe learning.
22 min read
How we built it
I Cut My Claude API Bill Without a Cloud Proxy — Here's How
A local-first way to skip repeat LLM calls, compress prompts 60–95%, and trigger your provider's own KV-cache discount — one line, open source.
5 min read
What's changing
EU AI Act, August 2: Where Does Your AI Agent's Memory Actually Go?
When the next phase of the EU AI Act applies, if your agent's memory runs through a cloud service you own a new question. A local-first, architecture-first take.
4 min read
What's changing
SLM Mesh v1.3.0 + SuperLocalMemory v3.4.48: Your AI Agents Now Work Across Machines
Multi-machine mesh support ships in both products. M4 and M5 coordinate as one. Real-time push, mDNS auto-discovery, zero config.
3 min read
Field reports
Severance for AI Agents: Your Coding Agent Is an Innie
Severance gave us a vocabulary for what AI coding agents actually do. They start every session as innies — no memory of yesterday's work. That is not a UX bug. It is the bottleneck.
7 min read
What's changing
Google Just Validated What We Built: Why Jitro Proves AI Agents Need Persistent Memory
Google's Project Jitro (Jules V2) is building a persistent agentic workspace with goals, insights, and history. This is exactly the problem SuperLocalMemory solved — locally, privately, and months earlier.
7 min read
Field reports
Why Every AI Coding Agent Will Need Persistent Memory by 2027
Open your terminal. Start a session with any major AI coding tool — Cursor, GitHub Copilot, Windsurf,...
7 min read
How we built it
I Gave Claude Code a Permanent Brain — Free, Local, 60 Seconds
Your AI agent forgets everything between sessions. Here's how to fix that with one command — no cloud, no API keys, no cost.
7 min read