Memory for Cline, Windsurf and ChatGPT: connecting MaxGraph in 10 minutes
Ready-made MCP configs for Cline, Windsurf (Devin Desktop) and ChatGPT: local MaxGraph server, access token, verification, common pitfalls.
OpenCode 2.0: the AI coding agent became a service — what changed
OpenCode 2.0 is a major overhaul of the most popular open-source coding agent (208K stars, 16M developers a month). The key change: the agent now runs as a persistent background service that the terminal, desktop app, and web UI all connect to. Plus hot-reloadable plugins, worktree tools, and 75+ model providers.
Liquid AI's LFM2.5-24B-A2B: 24B parameters, 2B active — agents move on-device
Liquid AI released LFM2.5-24B-A2B, a reasoning model with 24B parameters of which only 2B are active per token. It beats Qwen3.6-35B-A3B on five of eight agentic benchmarks, runs at 184 tokens/s on an Apple M5 Max, and is small enough to live on a laptop. Here's what it changes for local agents.
Apple's Minions: a GPU cloud built from idle Macs — how it works
Apple published Minions: an internal system that aggregates idle Apple Silicon GPUs across employees' Macs into a distributed inference cloud. GPU queue times dropped from 10+ hours to minutes, with zero hardware purchases. We break down the architecture from the paper.
Gemini 4 Argon: Google's new flagship — a million tokens and overnight agent runs
Gemini 4 Argon launched on October 13: Google's flagship reasoning model with a 1M-token context window. Pricing before and after December 1, availability gaps, and a close look at the headline case: an 11-hour unattended run over a 200,000-line codebase.
Why your AI agent forgets your project (and what to do about it)
AI agents forget projects for three reasons: the window resets between sessions, compaction cuts the middle, no state layer. Diagnose your case, pick a fix.
We measured the cost of re-explaining context: −43% tokens
Every new AI-agent session starts with re-explaining the project. We measured that ritual in tokens: memory cuts the starting context by 1.4–3.3×. Honest numbers, methodology, caveats.
The Model Context Protocol explained: a 2026 guide
MCP in plain English: why Anthropic created a single protocol for AI agent tools, how stdio differs from streamable HTTP, and how to connect your first server.
The agent memory glossary: 25 terms
25 agent-memory terms in plain definitions: context window, RAG, knowledge graph, embedding, consolidation, MCP, memory digest and more. Built to be quoted.
MaxGraph vs mem0 vs Zep: an honest comparison
MaxGraph, mem0 and Zep compared on seven axes: self-hosting, data storage, search, state, export, protocol, price. Including where competitors beat us.
Long-term memory for AI agents: what it is and how it works
Long-term memory for AI agents explained: why the context window isn't memory, how RAG, knowledge graphs and hybrids work, and how to choose the right approach.
Knowledge graph vs vector search for agent memory
An honest breakdown of two agent memory approaches: vectors win on paraphrases, graphs on state and relations. Comparison table and why we went hybrid.
How to save tokens when working with AI agents: 7 tactics with numbers
Tokens are eaten not by answers but by repetition: re-explaining the project, whole-file reads, bloated system prompts. 7 tactics to cut usage 1.4–4×, with measurements.
How to give Cursor memory between sessions
Set up long-term memory for Cursor in 10 minutes: MaxGraph in Docker, one mcp.json entry, a protocol block in project rules, and a write-read check.
How to give Claude Code memory between sessions
Set up long-term memory for Claude Code in 10 minutes: MaxGraph over MCP, a protocol block in CLAUDE.md, and a write-read check.
Context compaction: what it is and how to stop losing your work
Context compaction is when the client compresses conversation history into a summary. What gets lost, how to spot it, and how to work so it doesn't hurt.
Connecting MCP servers to your agent in 10 minutes
A hands-on MCP tutorial: where to find servers, how to connect them in Claude Code and Cursor, how to verify they work, and the six most common mistakes.