MaxGraph Blog

Memory for Cline, Windsurf and ChatGPT: connecting MaxGraph in 10 minutes

Fri, 16 Oct 2026 00:10:00 +0500

TL;DR: MaxGraph speaks MCP — the standard protocol — so it plugs into any compatible host. Below are ready-made configs for Cline, Windsurf (now Devin Desktop) and ChatGPT. We already have dedicated guides for Cursor and Claude Code.

What you need

A running local MaxGraph (5-minute install — one prompt to your agent)
Your access token: after the first launch it sits in data/local_token.txt
10 minutes

The endpoint is the same everywhere: http://localhost:8787/mcp, header Authorization: Bearer <your token>.

OpenCode 2.0: the AI coding agent became a service — what changed

Thu, 15 Oct 2026 19:00:00 +0500

TL;DR: Anomaly released OpenCode 2.0 — a major overhaul of the open-source AI coding agent (208,000 GitHub stars, 950 contributors, ~16M developers monthly). The central change is architectural: the agent now runs permanently as a single background service, and the terminal UI, desktop app, and web interface are just clients of it. Plugins are now hot-reloadable — the agent can write a plugin for itself mid-session and use it immediately.

What OpenCode is and why it needed a 2.0

OpenCode is an open-source agent that writes code in your terminal, IDE, or desktop app. It’s free, works with 75+ model providers through Models.dev (including local models), can sign in with your GitHub Copilot or ChatGPT Plus/Pro account, and doesn’t store your code on its servers.

Liquid AI's LFM2.5-24B-A2B: 24B parameters, 2B active — agents move on-device

Thu, 15 Oct 2026 12:00:00 +0500

TL;DR: On October 13, 2026, Liquid AI released LFM2.5-24B-A2B — a reasoning model with 24B parameters where only ~2B are active per token (Mixture-of-Experts). On five of eight agentic and tool-use benchmarks it beats Qwen3.6-35B-A3B, a model roughly 50% heavier in both total and active parameters. Speed: 184 tokens/s on an Apple M5 Max, and up to 21,000 output tokens/s on a single NVIDIA B300. Weights are open on Hugging Face — both the base and the post-trained versions.

Apple's Minions: a GPU cloud built from idle Macs — how it works

Wed, 14 Oct 2026 12:55:00 +0500

TL;DR: Apple published a paper called Minions (arXiv, October 2026): an internal system that pools idle GPUs on employees’ Macs into a distributed cloud for ML inference. Jobs that queued for centralized GPUs for 10+ hours now start within minutes — with zero new hardware. The architecture is deliberately pragmatic: PostgreSQL as the source of truth, no central scheduler, Ollama on every Mac.

The problem: GPUs queue for hours while hundreds of Macs sit idle

In any large company, ML inference lives on centralized GPU clusters: expensive, permanently oversubscribed, and jobs wait in line. The paper states the number honestly: some inference jobs waited in queue for over ten hours.

Gemini 4 Argon: Google's new flagship — a million tokens and overnight agent runs

Wed, 14 Oct 2026 00:30:00 +0500

TL;DR: On October 13, 2026, Google shipped Gemini 4 Argon (gemini-4-argon) — a new flagship reasoning model with a million-token context and a clear bet on autonomous agents. The headline launch case isn’t a greenfield demo: it’s an 11-hour unattended overnight run across a real 200,000-line repository. Access is limited and the rollout is bumpy; the introductory price of $2/$10 per million tokens doubles on December 1.

What the model is

From the model page in the Gemini API docs:

Connecting MCP servers to your agent in 10 minutes

Tue, 13 Oct 2026 00:30:00 +0500

Short version: An MCP server is a process that hands tools to your agent over a standard protocol. Connecting one comes down to a single config: a launch command for a stdio server, a URL and a token for an HTTP one. Below: where to find servers, ready-made configs for Claude Code and Cursor, how to verify everything works, and the six most common mistakes.

What you need

An MCP-capable client: Claude Code, Claude Desktop, Cursor, Windsurf or VS Code
For stdio servers: Node.js (npx) or Python (uvx) on your PATH
10 minutes

Context compaction: what it is and how to stop losing your work

Tue, 13 Oct 2026 00:30:00 +0500

Short version: Context compaction is what happens when your client hits the context window limit and replaces the conversation history with an automatic summary. The summary keeps the plot but drops the load-bearing details: exact wording of decisions, numbers, the “why”, and what’s left to do. To stop losing work, write important facts to external memory the moment they appear, close out stages explicitly, and don’t drag a session into an emergency compaction.

How to give Claude Code memory between sessions

Tue, 13 Oct 2026 00:30:00 +0500

Short version: Claude Code forgets your project when the session ends. Adding long-term memory takes ten minutes: run the local MaxGraph MCP server in Docker, connect it with one claude mcp add command, and paste a protocol block into CLAUDE.md. From then on, the agent records decisions, bugs and tasks — and restores them at the start of the next session.

What you need

Docker (Docker Desktop on Windows/macOS, docker + compose on Linux)
Claude Code
10 minutes

How to give Cursor memory between sessions

Tue, 13 Oct 2026 00:30:00 +0500

Short version: Cursor remembers your project only inside the current chat — a new session starts from zero. Adding long-term memory takes ten minutes: run the local MaxGraph MCP server in Docker, register it in mcp.json, and paste a protocol block into your project rules. From then on, the agent records decisions, bugs and tasks itself and restores them at the start of the next session.

What you need

Docker (Docker Desktop on Windows/macOS, docker + compose on Linux)
Cursor with MCP support (any recent version)
10 minutes

Step 1. Run the memory server

git clone https://github.com/RuMemory/maxgraph
cd maxgraph
docker compose up -d

Check:

How to save tokens when working with AI agents: 7 tactics with numbers

Tue, 13 Oct 2026 00:30:00 +0500

Short version: tokens are eaten not by the agent’s work but by repeatedly feeding it context. Seven tactics that work: memory instead of re-explaining the project, targeted search instead of whole files, a short system prompt, session discipline, matching the model to the task, prompt caching, and measuring usage. On our own repo this cut session-start tokens by 1.4–3.3×, and up to 4.2× on pinpoint questions.

Where the tokens actually go

Before saving, look at the anatomy of the bill. Every agent request pays for four buckets:

Knowledge graph vs vector search for agent memory

Tue, 13 Oct 2026 00:30:00 +0500

Short version: Vector search and knowledge graphs solve different memory problems. Vectors find “similar in meaning” — they save you when the question is phrased differently. A graph stores state and relations — it answers “what’s decided”, “what’s open”, “what depends on what”. Agent memory needs both properties, so the working answer is a hybrid: a graph as the core of state, semantic search as the way to find things in it.

Long-term memory for AI agents: what it is and how it works

Tue, 13 Oct 2026 00:30:00 +0500

Short version: long-term agent memory is a layer that preserves knowledge between sessions: decisions, project facts, preferences, open tasks. The context window is RAM that gets wiped; long-term memory is the disk. There are three ways to build it: document search (RAG), a knowledge graph, or a hybrid.

The context window is not memory

Context windows keep growing (200K, a million tokens), and the temptation follows: “just keep everything in the window”. Three reasons why that isn’t memory:

MaxGraph vs mem0 vs Zep: an honest comparison

Tue, 13 Oct 2026 00:30:00 +0500

Short version: mem0 is the most established option: managed cloud, ecosystem, benchmarks. Zep is the enterprise play with SOC 2 and HIPAA, but no self-hosted offering at a reasonable price. MaxGraph is a local memory server with tasks, bugs and sessions living in the same graph, for a micro-subscription. Below: a seven-axis table and an honest look at who’s stronger where, including our weak spots.

A note on method

We’re an interested party, so here are the rules: every fact about competitors comes from their public pages (pricing, READMEs, the GitHub REST API), snapshotted on October 10, 2026; the full research covering 18 memory systems is in the repo. Where we’re worse, we say so. Where a competitor’s number is marketing, we flag it.

The agent memory glossary: 25 terms

Tue, 13 Oct 2026 00:30:00 +0500

Short version: a dictionary of 25 terms without which any talk about AI-agent memory turns to fog: from context window to LLMO. Every definition is self-contained and can be quoted as is. Groups: foundations, storage, retrieval, protocols.

Foundations

Context window — the amount of text, in tokens, that a model sees at once: the system prompt, the conversation history, tool results. It’s the session’s RAM: a new session starts with an empty window.

The Model Context Protocol explained: a 2026 guide

Tue, 13 Oct 2026 00:30:00 +0500

Short version: MCP (Model Context Protocol) is an open protocol that Anthropic introduced in late 2024. It defines one standard way to plug external tools and data into an AI application: a server declares what it can do (tools, resources, prompts), and a client consumes that. Before MCP, every “model + service” pair was wired by hand; now a server is written once and works with every MCP client.

The problem MCP closed

Until November 2024, connecting an agent to the outside world looked like this: every service had its own API, and every client had its own function-calling format. M clients times N services make M×N integrations, each one hand-written. Want Claude to talk to GitHub — write a shim. Want the same for Cursor tomorrow — write a second one.

We measured the cost of re-explaining context: −43% tokens

Tue, 13 Oct 2026 00:30:00 +0500

TL;DR: a session with MaxGraph memory starts on ~4,200 tokens of context versus ~8,900 for reading the project from scratch — 2.1× smaller (honest range 1.4–3.3×). On pinpoint questions spanning two files — up to 4.2×. The headline says −43%: that’s the conservative floor, not the marketing maximum.

The ritual nobody counts

Open a new chat, explain the project again. The stack, the conventions, “yesterday we fixed the auth bug”, “deploys only go through the preflight script”. Anyone working with an agent beyond a single session does this daily. The ritual is so habitual that nobody measures its cost.

Why your AI agent forgets your project (and what to do about it)

Tue, 13 Oct 2026 00:30:00 +0500

Short version: Your agent doesn’t forget the project because the model is bad — it forgets because its memory is the context window, and nothing else. The window resets between sessions, compaction cuts it mid-session, and the default setup has no separate layer where project state could live. The fix is adding that layer — anything from a rules file to MCP memory; which one depends on what exactly you’re losing.