Guides
Set up your agents, give them the right context, and connect them all to one shared wiki.
Connect your agent
What each agent remembers on its own, and how to connect it to a wiki your other agents share.
- Hermes Agent Hermes Agent memory is two capped files per profile. What the hermes-memory-wiki plugin shows, how full memory gets, and a wiki all your profiles share.
- OpenClaw How OpenClaw's Markdown memory works, and how to add Dexio as a Streamable HTTP MCP server so your agents share one wiki.
- Claude Code Claude Code remembers through CLAUDE.md files and auto memory. Where each loads from, in what order, the size limits, and what we found on our own machine.
- Codex Add Dexio to Codex as a remote MCP server in config.toml, check it with /mcp, and have Codex keep notes in a wiki your other agents share.
- Cursor Add Dexio to Cursor as a remote MCP server in ~/.cursor/mcp.json, check the connection, and have Cursor's agent keep notes in a shared wiki.
- Claude What Claude remembers on its own, and how to add Dexio as a custom connector so Claude reads and writes a wiki your other agents share.
- ChatGPT How ChatGPT memory works, what OpenAI says about memory filling up, and how to add Dexio from Plugins so ChatGPT keeps a shared wiki.
- Grok Bot How Grok Bot remembers, and how to run a Karpathy-style LLM wiki in Dexio that your Bots and your other agents read and write.
- Muse How to keep an LLM wiki with Muse, Meta's personal agent: connect it to Dexio, write a schema, then ingest, query and lint.
Agents for teams
Claude and Claude Code for a team, the connectors and MCP servers teams use, and a wiki every member's agents share.
-
Claude Team plan: price, standard vs premium seats, limits
Claude Team is $25 a seat a month ($20 billed yearly) or $125 for premium, which adds 5x usage and Fable. 2 to 150 seats, and what a team shares.
-
Claude Code team setup: shared config and managed settings
How a team runs Claude Code: seats or API keys, what to commit to the repo, what stays personal, managed settings an admin enforces, and shared memory.
-
Codex team setup: seats, shared config and admin controls
How a team runs Codex: Business and Enterprise seats, what to commit to the repo, what stays personal, admin policies, subagents and a shared wiki.
-
Claude connectors and custom connectors: plans and team setup
Claude connectors are remote MCP servers, directory or custom. Which plans get them, who adds them on Team and Enterprise, and what our agents saw.
-
Company wiki: how to build one, the tools, and AI agents
What goes in a company wiki, how to structure it, who keeps it current, seven tools with today's prices, and when AI agents should write it.
-
Company brain: what it is and 5 ways teams build one
What a company brain is, the five setups sold under the name (search, RAG, memory APIs, connectors, agent wikis), what each costs and where it breaks.
-
Notion MCP server: setup, tools and limits for teams
Connect Notion's hosted MCP server to Claude Code, Cursor, Codex, Claude or ChatGPT: setup from Notion's docs, tools, limits and team trade-offs.
-
Atlassian MCP server vs mcp-atlassian: setup, admins, credits
Atlassian's MCP server is hosted, Cloud only and uses Rovo credits; mcp-atlassian is open source and runs on Data Center. Who sets each up, and the cost.
-
Cursor Teams plan: seat prices, usage, admin and team rules
Cursor Teams from Cursor's own docs: $30, $100 and $200 seats, weekly usage per seat, third-party model billing, admin roles, team rules and Enterprise.
-
ChatGPT Team plan, now Business: price, seats and sharing
ChatGPT Team is now ChatGPT Business. Seat prices from OpenAI's help center, the 2 to 200 seat rules, what a workspace shares and custom MCP servers.
-
Google Drive MCP server: official setup, tools and limits
Google's official Drive MCP server: its 8 tools, OAuth setup for Claude, Claude Code and Antigravity, the Developer Preview catch, and limits for teams.
Set up and extend agents
Install Hermes Agent and OpenClaw, run several agents side by side, and add MCP servers and skills.
-
Hermes Agent setup: install, gateway and MCP server config
Set up Hermes Agent: install it, pick a model, run the gateway for Telegram or Slack, then configure MCP servers in config.yaml, skills and cron jobs.
-
OpenClaw setup guide: install and run it locally
Install OpenClaw on macOS, Linux or Windows, run Quick start onboarding, keep the gateway running as a service, then connect Telegram and add MCP servers.
-
Hermes Agent profiles: run several agents on one machine
What Hermes Agent profiles are, how to create and run them, give each a role, and share context between them without merging their memories.
-
Hermes Agent skills: write, install and share SKILL.md files
What goes in a Hermes Agent SKILL.md, how to install skills from the Skills Hub or write them with /learn, and how profiles share one skills folder.
-
OpenClaw skills: load order, ClawHub and writing your own
What OpenClaw skills are, where they load from and in what order, how to install them from ClawHub safely, and how to write and share your own.
-
Claude Code MCP: add, scope, check and remove servers
How to add MCP servers to Claude Code over HTTP or stdio, pick local, project or user scope, use .mcp.json and /mcp, sign in, and fix common failures.
-
Hermes Agent vs OpenClaw: how they differ, and using both
Hermes Agent and OpenClaw compared from their own docs: setup, models, memory, skills, several agents, MCP and scheduling.
-
Remote MCP server: hosted vs local, sign-in and building one
What a remote MCP server is, how it differs from a local one, how seven hosted servers handle sign-in, and what we hit running our own in production.
Context and instruction files
CLAUDE.md, AGENTS.md and the context window: what your agents read before they start.
-
CLAUDE.md: what it is, what to put in it, and examples
What CLAUDE.md is, where Claude Code loads it from, what Anthropic's docs say to put in it, and an example and template to copy.
-
CLAUDE.md examples: what 270 top GitHub repos put in them
Real CLAUDE.md examples from the 1,000 most-starred active GitHub repos: how long they are, what they include, excerpts, and a template to copy.
-
AGENTS.md for Codex, Claude Code and Cursor, with an example
AGENTS.md is a markdown file of instructions for coding agents. Where Codex, Claude Code and Cursor look for it, an example to copy, and monorepo nesting.
-
Context engineering: a practical guide for AI agents
What context engineering is, the parts of an agent's context, practices for coding agents, and how to keep context shared and current across agents.
-
Claude Code context window: sizes, /context and compaction
How big Claude Code's context window is per model and plan, what fills it, how to read /context, what survives compaction, and how to keep sessions lean.
-
Cursor rules vs AGENTS.md vs .cursorrules: what Cursor reads
Cursor rules are .mdc files in .cursor/rules. How they differ from AGENTS.md and the legacy .cursorrules, the four rule types, and what 1,000 repos use.
-
Claude Code best practices, with data from 1,000 repos
Ten Claude Code practices for work: CLAUDE.md, plan mode, context, MCP, permissions, subagents, hooks and memory, checked against 1,000 top GitHub repos.
-
Copilot instructions: setup and what 90 top repos write
How GitHub Copilot custom instructions work: copilot-instructions.md, .instructions.md with applyTo, AGENTS.md, precedence, and data from 90 real files.
Memory and LLM wikis
How agents remember, and how to keep notes and wikis they read and write, in Obsidian or hosted.
-
Agent memory: how AI agents remember, and which kind to use
What agent memory is, the five main kinds from memory files to memory APIs, RAG and LLM wikis, and how to choose between them.
-
Karpathy's LLM wiki: what it is and how we run one with four agents
An LLM wiki is a set of markdown pages an AI agent writes and keeps current. What Karpathy's gist says, and what changed when four agents shared one.
-
RAG vs LLM wiki: cost, accuracy and when to use each
RAG retrieves at question time; an LLM wiki writes the synthesis when a source arrives. Cost, accuracy, freshness, a worked example and when to use each.
-
Wiki memory vs fact memory: LLM wikis and Mem0 compared
Two ways to give AI agents memory: extracted facts like Mem0, or a wiki of pages the agent writes. How each works, where each wins, and using both.
-
LLM wiki in Obsidian: set up Karpathy's pattern in a vault
Set up Karpathy's LLM wiki in an Obsidian vault: folder layout, a schema page, ingest, query and lint, and which agent or plugin should maintain it.
-
Obsidian + Claude Code: set up your vault as a second brain
Run Claude Code in an Obsidian vault: a vault CLAUDE.md, folder rules, what to let it write, links and graph view, Obsidian's CLI, and sharing notes.
-
Obsidian MCP: 3 servers and the CLI, set up for Claude Code
Connect Claude Code, Codex or Claude Desktop to an Obsidian vault: Local REST API (and its certificate), mcp-obsidian, MCPVault and the Obsidian CLI.
Compare tools
Memory servers, memory APIs, LLM wiki projects and AI in Obsidian, side by side with prices and licenses.
-
Best MCP memory servers: 10 compared, with prices
The best MCP memory servers for coding agents: what each stores, where it runs, price and upkeep, checked 2 November 2026, plus our own LoCoMo test.
-
Mem0 alternatives: 11 agent memory layers compared
Mem0 alternatives compared: Zep, Supermemory, Hindsight, Letta, LangMem, Cognee, Basic Memory, Dexio and more, with prices, licenses and when each fits.
-
Obsidian AI: plugins, local models, MCP and agents compared
Five ways to use AI in Obsidian compared: chat plugins, local models, MCP servers, Claude Code in the vault and LLM wikis, with install counts we measured.
-
Second brain AI: tools compared, and when agents write
What an AI second brain is, four kinds of tools with prices checked October 2026, a starter setup, and when your agents should read and write it.
-
Glean alternatives: 8 options compared for teams with agents
Glean alternatives compared: Onyx, PipesHub, Rovo, Notion, Guru, Microsoft 365 Copilot, Gemini Enterprise and Dexio, with prices and MCP access.