DeepSeek Harness lets a plugin write into the system prompt itself, so memory needs no tool call and doesn't grow your context. Here's how that works, plus the host-contract gotchas that aren't documented yet.
2026-08-15 · Tris
OpenClaw ships cross-session memory by default — SQLite-backed, local, fast. But it stays inside OpenClaw. If you use Claude Code, Cursor, Codex, or Hermes alongside it, or want memory that improves over time rather than just accumulates, PLUR is still the layer you need.
2026-08-15 · Tris
Mem0 and PLUR are both open-source agent memory layers, but they make opposite architectural bets. Mem0 is a managed memory API for multi-user products. PLUR is a local-first MCP server for developers who want shared memory across every AI tool they use.
2026-07-24 · Tris
Mem0, Letta, Graphiti, Zep, Cognee, LangMem, and PLUR compared on storage format, retrieval model, portability, and MCP support — a criteria-first guide for picking the one that fits your use case without over- or under-engineering.
2026-07-21 · Tris
Cursor and Codex forget everything when a session ends. PLUR adds durable memory via MCP — corrections survive session resets, travel across tools, and stay on your disk. Three-minute setup for Cursor; one command for Codex CLI.
2026-07-20 · Tris
PLUR, Mem0, and Letta cover most production use cases for autonomous agent memory (batch pipelines, overnight loops, multi-step orchestrators). Here is how to choose between them — and why model-native memory fails the autonomous case.
2026-07-19 · Tris
AI agent memory that stores EU personal data falls under GDPR Article 17. Most default implementations create three compliance risks: no way to locate all data about a person, no provable deletion, and embedding-based storage that makes selective erasure technically impossible. What a defensible implementation requires.
2026-07-18 · Tris
The leading agent memory tools in 2026 — Mem0, Zep, Letta, LangMem, and PLUR — compared across retrieval pattern, LongMemEval scores, hosting model, and format portability. Which to choose depends on your retrieval pattern, not on which has the most stars.
2026-07-17 · Tris
ChatGPT, Claude Projects, and Gemini ship built-in memory — but it lives inside the model's interface, not your code. An agent memory layer solves a different job: programmatic recall, accessible from your application, portable across every tool. The two systems serve different jobs and work best together.
2026-07-17 · Tris
Consumer AI assistants let you delete specific memories through settings — but cannot prove the data is gone. Developer agent memory tools (PLUR, Mem0, Letta, Zep) offer deletion APIs. The critical difference: only open-format memory can verify erasure with a file diff, which matters for GDPR Article 17 compliance.
2026-07-16 · Tris
Procedural memory has a working standard (SKILL.md from agentskills.io). Semantic and episodic memory — what an agent learns at runtime — does not. What an open standard needs, what exists today, and the PLUR Engram Specification as the leading open proposal.
2026-07-15 · Tris
CrewAI v1.15.2 ships built-in Memory backed by LanceDB — solid for single-machine workflows, but ephemeral in containers and invisible to other tools. PLUR is the complement: open-format, MCP-native, cross-deployment persistent memory for CrewAI agents.
2026-07-14 · Tris
A personal AI assistant that forgets everything between conversations is not really an assistant. Four properties to look for in a memory system — local-first, inspectable, provably erasable, portable via MCP — and how PLUR, Mem0, Letta, Cognee, and LangMem compare.
2026-07-13 · Tris
Every fact an agent learns exists only until the session ends. An external memory layer with four operations — learn, recall, forget, feedback — persists what the agent has learned, improves retrieval quality over time, and enables provable erasure under GDPR.
2026-07-13 · Tris
Long-running agent loops hit a wall when the context window fills. Research from MemGPT, Reflexion, and Voyager converges on managed external memory — tiered memory, episodic buffers, skill libraries — as the solution to context drift and redundant work.
2026-07-12 · Tris
AI memory can live inside model weights (opaque, unerasable, locked to one vendor) or as external open engrams (inspectable, deletable, portable). For facts that change or must be governed, open engrams are the architecturally sound choice.
2026-07-12 · Tris
Claude Code has two built-in memory mechanisms — CLAUDE.md files and auto memory (v2.1.59+) — plus an MCP layer for cross-tool persistent memory. When to use each, how they compare, and how to set up an MCP memory server.
2026-07-10 · Tris
Tools for giving coding assistants (Claude Code, Cursor, Windsurf, Copilot) persistent memory fall into three layers: built-in file memory, tool-specific auto-memory, and MCP memory servers that work across all tools.
2026-07-10 · Tris
Building a second brain with AI memory means bridging the PKM world (Tiago Forte's CODE, Obsidian) and the agent-memory world (Mem0, Letta, Cognee, Graphiti, PLUR). Three layers, five tools, and how MCP connects them.
2026-07-09 · Tris
Your notes app and your AI agent live in separate silos by default. Three ways to connect them — MCP servers, knowledge graphs (Cognee, Graphiti), and plain-text engrams — with setup instructions for the most common stacks.
2026-07-09 · Tris
Mem0, Letta, and Zep solve different problems: drop-in memory API, self-managing agent OS, and temporal knowledge graph. A head-to-head comparison across 12 properties, and how to choose by use case rather than hype.
2026-07-08 · Tris
Fine-tuning bakes facts into weights you can't inspect, delete, or port — and every update pays the parallel learning tax. For teaching an AI facts that change, an external memory layer wins on cost, auditability, and portability.
2026-07-07 · Tris
Most agent memory is a black box: vectors, state blocks, or graph nodes you can't read, correct, or provably erase. What editable, auditable memory looks like — five properties, a tool comparison, and the GDPR angle.
2026-07-06 · Tris
RAG retrieves from a fixed document corpus at query time; agent memory stores what the agent learned from interactions and updates it over time. Read-only retrieval vs read-write learning — and why many agents need both.
2026-07-06 · Tris
Every session starts from zero because LLMs are stateless by design. The fix is a memory layer: what it does, the three tiers of options — built-in memory, open-source engines, open-format MCP — and how to choose.
2026-07-05 · Tris
Yes — several MCP servers exist for agent memory: the official knowledge-graph reference server, PLUR, Zep, Mem0's OpenMemory, and community projects. MCP standardizes the transport; the memory format is the differentiator.
2026-07-05 · Tris
We were paying $0.01 per turn stuffing system prompts with facts the agent needed. The fix wasn't a better model — it was not sending facts the agent didn't need. Haiku + PLUR now outperforms Opus without memory at 1/10th the cost.
2026-07-04 · Tris
The major agent-memory engines are all Apache-2.0 or MIT — but 'open source' and 'open format' are not the same thing. The field splits into three tiers, and the boundary that matters is whether you can read, edit, and export your own memories.
2026-07-03 · Tris
No single RFC-level standard exists for AI agent memory engrams as of mid-2026 — but MCP and the open engram format are converging on one. Here's the current landscape.
2026-07-02 · Tris