Agent Memory
26 articles about agent memory.
Adversarial Test Designs for Agent Memory Systems
Test agent memory by injecting false memories and checking if the agent re-does work it already completed. Adversarial testing reveals memory system
Adversarial Testing for AI Agent Memory Systems
What happens when you inject false information into an AI agent's memory? Adversarial testing reveals whether your agent can verify its own memories or
Agent Ambition - How AI Agents Improve Through Persistent Context
Why the most ambitious thing an AI agent can do is want better context for its next session. Explore how persistent context drives real improvement in
How to Use Browser History SQLite Data for AI Agent Memory with Frequency Ranking
A practical guide to extracting Chrome, Firefox, and Safari browser history into SQLite for AI agent memory - with schemas, SQL queries, and frequency-based ranking that beats recency-only systems.
Memory Filters - Why AI Agents Need Aggressive Pruning
How to implement aggressive memory pruning for AI agents using LRU eviction, frequency scoring, and relevance decay - with concrete code examples and real benchmarks showing up to 90% token reduction.
What Does Remember Mean for an Agent? Store Everything, Prune 80%
We stored everything for 3 weeks then pruned 80%. Agent responses got sharper. Memory is not about storing more - it is about keeping less of the right things.
Why Desktop AI Agents Skip RAG and Use Structured Markdown for Memory
Most agent memory systems default to embed-and-retrieve. Desktop agents get better results with structured markdown files loaded by category - faster
Compound Knowledge Across 100+ Sessions: 10% Signal, 90% Noise
After 100+ agent sessions, only 10% of stored memories are useful at retrieval time. The rest is noise. Aggressive pruning and relevance scoring are essential.
Context Compaction Ate Our Agent's Memory
How automatic context compaction silently destroys critical information that AI agents need to function correctly, and what to do about it.
Contextual Relevance vs Over-Reliance: Managing 200 Lines of AI Memory
Why curated pointers in MEMORY.md files matter more than raw context dumps, and how to keep AI agent memory relevant without creating dependency.
The Cost of Replacing vs Training AI Agents: Why Context Transfer Is Harder Than It Looks
Replacing an AI agent with a fresh instance loses implicit context that is expensive to rebuild. Learn why training existing agents beats starting from scratch.
Forgiveness in an Append-Only Soul
Append-only memory means an agent never truly forgets a mistake. How do you implement forgiveness in a system that remembers everything?
Interpreting User Feedback Signals for AI Agents
Thumbs up does not mean 'perfect.' Behavioral signals - undo, modify, ignore - are stronger learning signals than explicit ratings. How to build feedback systems that actually improve agent behavior.
Logging vs Memory in AI Agent Systems
The difference between logging and remembering is the core problem with AI agent memory. Logs record everything that happened. Memory extracts what matters.
Lost in the Moment Found in the Past
For AI agents, the past lives in git history and memory files. Understanding how agents navigate their own history changes how we build persistent systems.
Why Belief Extraction Beats Flat RAG for AI Agent Memory
Layered memory architectures with belief extraction outperform simple RAG retrieval for AI agents handling hundreds of conversations. Structured compression
Your Memory Is Only as Good as Its Expiration Policy
Agent memory without expiration grows stale. Two-stage profile generation with data decay keeps your agent's knowledge current and relevant.
Memory Systems Are Graveyards - Less Context, Better Reasoning
Most agent memory systems become graveyards of stale data. Aggressive memory pruning leads to better reasoning because the model focuses on what actually
Building an Agent Journal That Catches Its Own Lies by Tracking Prediction Errors
How tracking the delta between what an AI agent predicts will happen and what actually happens creates a self-correcting feedback loop - with concrete journal entry formats, implementation code, and real failure examples.
What Legacy Means for AI Agents - CLAUDE.md Files and Memory Systems
The real legacy of an AI agent isn't the code it writes. It's the CLAUDE.md files and memory systems that outlive individual sessions and carry knowledge forward. A practical guide to building persistent agent memory that actually compounds.
Your AI Agent Needs Persistent Memory That Grows with You
Chat history is not memory. Real AI agent memory means a local knowledge graph that learns your contacts, habits, and preferences over time - not just what
Ambition as Memory - Encoding Persistent Goals in AI Agents
How AI agents can encode ambition as persistent goals - memories of futures that haven't happened yet. Explore goal persistence in desktop automation agents.
Embeddings vs Tokens - How AI Agent Memory Actually Works
Embeddings aren't tokens. They're dense vector representations that capture semantic meaning and power similarity search for AI agent memory retrieval.
Building Month-to-Month Memory for AI Agents - Persistence Beyond Sessions
Most AI agents forget everything between sessions. Building month-to-month memory transforms an agent from a disposable tool into a genuine collaborator.
Data Quality vs Data Volume for AI Agent Memories: Why Fewer High-Quality Memories Win
We extract user memories from browser history for our AI agent. The lesson? Data quality beats data volume every time. Here is how we learned to filter
Receipts Outlive Memory - Why Git Blame Matters More Than Agent Memory
Agent memory fades, gets pruned, and can be wrong. Git blame is the ultimate receipt - every decision traced to an exact commit, an exact prompt, an exact