Questions about AI memory and context, answered

Short, sourced answers to the questions teams ask about Claude, ChatGPT, Copilot, Cursor, Grok and shared AI memory.

Claude for teams

  • How do Claude Projects work, and what are they good for?

    A Claude Project is a workspace with its own chats, knowledge files, custom instructions and memory; each chat uses that knowledge and those instructions, but not the other chats. Free allows up to 5 Projects; paid plans have unlimited Projects and switch to retrieval (RAG) for up to ten times more capacity when knowledge nears the context limit. Sharing is limited to Team and Enterprise.

  • Does Claude have memory across chats, and how does it work?

    Yes. Claude saves useful details from your chats as individual memory topics on Free, Pro, Max, Ultra, Team and Enterprise, and each Project keeps its own separate memory. On Free and individual paid plans memory is on by default; on Team and Enterprise the organization owner decides whether it is available, and searching past chats is a separate feature limited to paid plans.

  • How do I import my ChatGPT memory into Claude?

    Copy Anthropic's import prompt from claude.com/import-memory, paste it into ChatGPT, and it asks for your saved memories and learned context in one code block. Remove anything sensitive, then in Claude go to Settings > Your Claude > Start import, paste the text and click Add to memory. It works on every Claude plan on web and desktop, without moving your chat history.

  • How do you use Claude Projects effectively?

    Use a Project for context that keeps coming back: one client, one product, one recurring report. Put the rules that must always apply in the project instructions, keep the knowledge files focused and named clearly, start a new chat for each task, and decide per Project whether it should share account memory. Content in a Project is cached, so reuse also counts less against your usage.

  • Is there a memory plugin for Claude, and which kind do you need?

    Yes, in three forms. A custom connector links Claude to a remote MCP memory server and works in Claude, Cowork and Claude Desktop on every plan (Free: one connector). A desktop extension runs a local MCP memory server on your own computer. A plugin bundles skills, connectors and sub-agents into one install. None of them changes Claude's built-in memory; they give Claude tools to read and write memory elsewhere.

  • What are Claude Cowork projects, and how do they differ from chat Projects?

    A Cowork project is a workspace for tasks: it has its own instructions, scheduled tasks, linked context (a local folder, a chat Project or a URL) and its own memory. A chat Project groups conversations around uploaded knowledge and instructions. You can import a chat Project into Cowork, share Cowork projects on Team and Enterprise, and keep memory separate per project. Cowork projects are not available in Claude Code yet.

All 22 questions →

Claude Code

  • How much does Claude Code cost, and which plan do I need?

    Claude Code is available through paid Claude plans or pay-as-you-go API access; the Free plan does not include it. As of November 2026, Pro costs $20 a month ($17 billed yearly), Max $100 and Ultra $200. Team Standard seats cost $25 a month or $20 billed yearly; API spending depends on usage.

  • How do I install Claude Code, and how do I keep it updated?

    Run the native installer: curl -fsSL https://claude.ai/install.sh | bash on macOS, Linux and WSL, or irm https://claude.ai/install.ps1 | iex in Windows PowerShell. Then run claude in your project and log in with a paid Claude plan or a Console account. Native installs update themselves, npm installs too when the npm folder is writable; Homebrew and WinGet need manual upgrades by default.

  • What are Claude Code skills and how do you write one?

    A skill is a folder with a SKILL.md file: a short YAML header with a name and description, then instructions in Markdown. Claude sees only the name and description in every session and reads the full file when a task fits, or when you type /skill-name. Personal skills live in ~/.claude/skills/, project skills in .claude/skills/ in the repo, and plugins can ship their own.

  • Claude Code desktop, CLI, web or mobile: what is the difference?

    All four use the same Claude Code engine, but cloud sessions load only part of your configuration. The terminal CLI runs on your machine and is the most complete. The desktop app wraps it in a visual interface with diff review and parallel sessions. Claude Code on the web runs sessions on Anthropic-managed cloud machines. The mobile app runs nothing itself: it steers cloud or local sessions.

  • How do I use Claude Code well? A first workflow that works

    Write a specific prompt, give Claude a check it can run (a test, a build, a linter) and ask for the output as evidence. Keep CLAUDE.md short and only for what the code cannot tell Claude. Put rules that may never be broken in deny rules or hooks, not in prose. Press Esc when it drifts, and after two failed corrections run /clear and write a better prompt.

  • How do I use Claude Code with GitHub, and @claude in pull requests?

    Claude Code GitHub Actions is a GitHub Action, anthropics/claude-code-action@v1, that runs Claude Code inside your repository's workflows. The quick setup is /install-github-app in Claude Code, which needs admin access to the repo. After that, anyone with write access can mention @claude in an issue or PR to get code analyzed, changed and pushed. Each run uses GitHub Actions minutes plus API tokens or your Claude subscription.

All 24 questions →

ChatGPT for business

  • ChatGPT for small business: how do you set it up for a small team?

    For a small business, the sensible setup is one ChatGPT Business workspace instead of personal Plus accounts: from $20 per Standard seat a month billed annually, usually with a two-seat minimum. Give heavy users a Premium seat, put recurring work in shared projects, decide your memory and data rules up front, and plan around custom GPTs being retired in favor of plugins on December 11, 2026.

  • How much does ChatGPT Team (Business) cost per user?

    As of November 2026, ChatGPT Business (formerly ChatGPT Team) costs $25 per Standard seat per month, or $20 with annual billing; Premium costs $125 or $100 respectively. A workspace usually needs 2 to 200 paid seats (some new workspaces can start with 1), mixing both types. Prices vary by country and currency, and Enterprise pricing is available through sales.

  • What are ChatGPT Projects and how do you use them for work?

    A ChatGPT project is a workspace for one topic: it keeps related chats, uploaded files, app links and project instructions together, and ChatGPT uses that context in every chat inside it. Projects are included on all plans, you can create as many as you like, and each holds 5 to 40 files depending on the plan. Project instructions override your global custom instructions.

  • How do ChatGPT custom instructions work, and what should you put in them for work?

    Custom instructions are a standing note to ChatGPT about who you are and how you want answers, applied to all your chats. Set them under Settings > Personalization with Enable customization on. Free and Go allow 1,500 characters, paid plans including Business 5,000. Project instructions override them inside a project, and custom GPTs ignore them. They stay personal: colleagues do not see or share them.

  • How does ChatGPT memory work?

    When Memory is on, ChatGPT can personalize answers with saved memories, relevant past chats, custom instructions, saved files and, depending on plan and region, connected apps such as Gmail. It does not keep every detail: it picks what seems relevant. You review and change it under Settings > Personalization > Memory, and it always belongs to one account.

  • Is there a free trial for ChatGPT Business?

    OpenAI's help articles describe no standard free trial for ChatGPT Business. Signing up is a paid checkout. Some customers can start with one seat; Standard costs $25 per user per month, or $20 billed annually, per November 2026. Promotional offers have appeared at checkout but are not documented as a fixed trial. The documented promotion is a $50 workspace credit for using certain partner plugins.

All 24 questions →

Microsoft Copilot

  • How does Copilot Studio pricing work: credits, capacity packs and pay-as-you-go?

    Copilot Studio bills in Copilot Credits: as of November 2026 you can buy a tenant pack of 25,000 credits for $200 a month, pay as you go at $0.01 per credit through an Azure subscription, or prepurchase a year of credit commit units for up to 20% off. Some agent use in Copilot Chat, Teams or SharePoint by users with a Microsoft 365 Copilot license is not billed separately.

  • What are Copilot agents and which ones come with Microsoft 365 Copilot?

    Copilot agents are scoped or focused versions of Microsoft Copilot that act as AI assistants and can automate business processes. Copilot Chat includes agents grounded in instructions and public websites; agents that use work data are pay-as-you-go. The paid Microsoft 365 Copilot license includes agents on web and work data, SharePoint agents, and prebuilt agents for research and data analysis; Microsoft names Researcher, Analyst and Facilitator.

  • What is the difference between Copilot Chat and a paid Microsoft 365 Copilot license?

    Copilot Chat is included at no extra cost with eligible Microsoft 365 licenses and answers from web data; it only uses work data you upload, open content in Outlook or Teams, or a pay-as-you-go agent. The paid Microsoft 365 Copilot add-on grounds answers automatically in your email, files, calendar and meetings through Microsoft Graph and Work IQ, with priority access in the Office apps.

  • How much does Microsoft 365 Copilot cost per user for a small business?

    As of November 2026, the Microsoft 365 Copilot Business add-on costs $18 per user per month on a yearly plan, a promotional price that is normally $21, or $25.20 billed monthly. Enterprise customers pay $30 per user per month billed yearly. Copilot Chat with web data is included at no extra cost with eligible Microsoft 365 business plans.

  • What license do you need for Copilot Studio?

    To build in Copilot Studio a person needs one of four things: the user license (which only works once the tenant has a prepaid Copilot Credit pack), membership of the Copilot Studio authors group for pay-as-you-go, a Microsoft 365 Copilot license, or a trial. The license gives access; running agents costs Copilot Credits.

  • What is the difference between Microsoft Copilot and Copilot Studio?

    Microsoft Copilot is the AI assistant people use in the Copilot app, Teams, Outlook, Word and Excel. Copilot Studio is a separate low-code studio at copilotstudio.microsoft.com where makers build agents and workflows, connect them to data and systems, and publish them to Copilot, Teams, websites or mobile apps. You use Copilot; you build with Copilot Studio.

All 18 questions →

Cursor

  • How do you use Cursor rules (.cursor/rules) and which rule type should you pick?

    Project rules are .mdc files in .cursor/rules, kept in version control. The description, globs, and alwaysApply fields set one of four modes: Always Apply, Apply Intelligently, Apply to Specific Files, or Apply Manually. Cursor ignores a plain .md file in that folder; use AGENTS.md for plain markdown.

  • Did Cursor remove Memories, and how do you keep memory across sessions now?

    Yes. Cursor removed the old Memories feature in version 2.1.x, according to a November 2025 staff reply. That does not mean Cursor has no persistent memory today: Automations can learn from earlier runs, and Projects retain learned context. For explicit project instructions, use Project Rules or AGENTS.md.

  • Cursor rules vs skills: what is the difference and when do you use each?

    A Cursor rule is an instruction that goes into the agent's context: always, for matching files, when relevant, or on @-mention. A skill is a portable folder with a SKILL.md and optional scripts, references, and assets that the agent loads when it judges it relevant, or that you call with /skill-name. Use rules for standing agreements, skills for procedures.

  • What makes a good Cursor rule? Best practices and examples

    Cursor's own guidance: keep each rule focused, actionable and scoped, under 500 lines, with concrete examples or references to files in your codebase. Do not paste style guides a linter can enforce or commands Agent already knows. Start small and add a rule only when Agent makes the same mistake twice. Community collections are a source of ideas, not a set to copy whole.

  • How do you set Cursor rules that apply to every project?

    Set them as User Rules in Customize > Rules. User Rules are stored on your Cursor account, sync to other machines when you sign in, apply across all projects, and are used by Agent (Chat) only, not by Tab or Bugbot. Files in ~/.cursor/rules stay on the local machine. When rules conflict, Team Rules win, then Project Rules, then User Rules.

  • How do you share context and team rules across a team in Cursor?

    Share explicit instructions through repository files or Team Rules, and tools through Team MCP servers. For learned context, Multiplayer Projects share a Project store among members; Teams and Enterprise also support shared chat transcripts. Personal preferences stay separate unless you deliberately publish or copy them into a shared place.

All 10 questions →

Grok for business

  • What does the Grok API cost per million tokens?

    As of November 2026, xAI's flagship grok-4.7 costs $2.00 per million input tokens, $0.50 for cached input and $6.00 per million output tokens. The cheaper grok-4.3 and grok-4.20 models cost $1.25 in and $2.50 out with a 1M context window. Long-context rates apply from 200k prompt tokens; the listed grok-4.x models receive a 20 percent batch discount.

  • What is Grok Business, and how do you set it up for a team?

    Grok Business is the self-serve team plan of xAI (now signing as SpaceXSI), listed at $30 per user per month as of November 2026. It gives each member a team workspace on grok.com next to their personal one, under business terms with no training on your data. An admin runs it from console.x.ai: buying licenses, inviting people, adding connectors and setting how widely members may share.

  • What is Grok Enterprise and how is it different from Grok Business?

    Grok Enterprise is xAI's top plan for organizations, sold through sales. It adds custom SSO, Directory Sync (SCIM), custom role-based access and advanced user management to Business. Only Enterprise customers can add Enterprise Vault, a dedicated data plane with application-level encryption and customer-managed keys.

  • Can a business use Grok Bot as a shared bot for the whole team?

    Yes. On Team and Enterprise plans, the owner of a Grok Bot can publish it as a Team Bot and share it with the team, the organization or chosen people. Everyone talks to the same bot, with the same apps, skills and team memory, but each chat is private. What a teammate teaches it for the team waits until the owner or a Manager approves it under Review learnings.

  • How much does Grok Business and Grok Enterprise cost for a team?

    As of November 2026, Grok Business costs $30 per user per month before applicable taxes and is available through self-service. Grok Enterprise has no public price: xAI quotes it through sales, and the Enterprise Vault add-on is quoted on top. Developer use of the Grok API is a separate, usage-based bill per million tokens.

  • Does Grok have memory like ChatGPT, and is it shared across a team?

    Grok has memory features, but the product matters: Grok Build keeps project notes across sessions, and Grok Bot Team Bots share approved team memory. These are different from personal chat memory in grok.com. The official pages checked do not establish that a Business or Enterprise workspace automatically shares each user's personal chat memory with colleagues.

All 10 questions →

Gemini and NotebookLM

  • How much does NotebookLM cost for a team?

    NotebookLM, renamed Gemini Notebook in July 2026, is free with a Google account. Higher limits come through a Google AI plan (Plus, Pro, Ultra), a Workspace edition from Business Standard up, or Gemini Notebook Enterprise in Google Cloud, sold per license with at least 15 licenses. For a team, the Workspace or Cloud route matters most because of the data terms.

  • Does NotebookLM have an API?

    Yes, two official ones, both in preview. The Gemini Notebook API, for personal Google accounts only, creates notebooks and adds sources, and is free during the preview. The Gemini Notebook Enterprise API in Google Cloud manages notebooks and sharing for licensed users. Neither documents a way to ask a notebook a question. Libraries such as notebooklm-py do more, but use undocumented Google endpoints.

  • Can you connect NotebookLM to Claude or Cursor with MCP?

    Yes, in two ways. Google runs an official remote MCP server for Gemini Notebook, the new name of NotebookLM, at notebook.googleapis.com/mcp, with setup steps for Claude Code and Cursor. It is a preview for personal Google accounts and its seven tools create notebooks and add sources, not ask questions. Community MCP servers can query notebooks, but log in with your browser cookies.

  • Is a Gemini notebook the same as NotebookLM?

    Mostly, yes. Google renamed NotebookLM to Gemini Notebook on July 16, 2026, and notebooks you create in the Gemini app sync both ways with the standalone Gemini Notebook app once it is connected. The notebook is the same; the room differs. The Gemini app adds web search and Gemini's tools to an ongoing chat; the standalone app is the research workspace with its own outputs.

  • What are NotebookLM's limits and limitations?

    In NotebookLM, now Gemini Notebook, the free plan allows 100 notebooks with 50 sources each; Pro allows 500 notebooks with 300 sources; Ultra up to 600 sources. Each source can hold up to 500,000 words or 200 MB. The bigger limitations are structural: notebooks cannot see each other, and the enterprise version does not save chats.

  • What is NotebookLM Enterprise, and how is it different from NotebookLM in Workspace?

    NotebookLM Enterprise, renamed Gemini Notebook Enterprise, is the Google Cloud version of NotebookLM. It runs in your own Cloud project, stores sources in the US or EU multi-region, supports VPC Service Controls, customer-managed keys and identity providers like Entra ID or Okta, and is licensed per user. The Workspace version is simpler to switch on, but data region settings do not apply to it.

All 14 questions →

Codex

  • What does a good AGENTS.md for Codex look like? An example

    A good AGENTS.md for Codex is short and specific: how to build, test and lint, where things live, the conventions Codex cannot guess, the things it must never do, and what done means. OpenAI's own guidance is that a short, accurate file beats a long one full of vague rules. Put rules that only apply to one part of a monorepo in an AGENTS.md in that directory.

  • How much does Codex cost for a team?

    For a team, Codex comes with ChatGPT Business: as of November 2026, $20 per user per month billed annually, or $25 per user billed monthly, with a minimum of two users. Standard seats come with included usage, estimated per five hours, and weekly limits may also apply; extra usage is bought as workspace credits. Enterprise and Edu are priced by sales, and Codex on an API key is paid per token.

  • Codex vs Claude Code: which should a team use?

    For a team, Codex (ChatGPT Business) and Claude Code (Claude Team) both cost $20 per user per month billed annually or $25 monthly as of November 2026. Both run in the terminal, IDE, desktop and cloud, both speak MCP, and both can take instructions from one AGENTS.md. Start with the tool your existing subscription already includes, then try both on the same task in your own repo; neither tool's local memory pools what teammates learn.

  • Does Codex remember previous work, and can a team share that memory?

    Codex can remember earlier chats through Codex memories, but they are off by default. Once on, Codex turns earlier chats into summaries and stores them as local files under ~/.codex on your machine. Team sharing is not part of it; OpenAI itself says required team guidance belongs in AGENTS.md or checked-in documentation, not in memories.

  • What is AGENTS.md and how does Codex read it?

    AGENTS.md is a plain Markdown file of instructions for coding agents: a README for agents, with no required fields. Codex first reads one global file from ~/.codex, then at most one AGENTS.md per directory from the project root down to your working directory. It joins them in order, the closest file last, and stops at 32 KiB by default.

Context engineering

  • What is a context graph?

    In Foundation Capital's definition, a context graph is a structure of decision traces linked across entities and time. A decision trace records what happened in one concrete case: the inputs from several systems, the policy applied, the exception granted and who approved it. Linked together, those traces make precedent searchable, so an AI agent sees not just the current state but why it is that way.

  • What is agentic context engineering (ACE)?

    Agentic Context Engineering (ACE) is a method from a 2025 paper by researchers including Qizheng Zhang, James Zou and Kunle Olukotun. It improves an AI agent by letting its context evolve into a playbook through three roles, generation, reflection and curation, with small incremental updates instead of rewriting. The abstract reports gains of about 10.6% on agent benchmarks and 8.6% on finance tasks, without changing model weights.

  • What is the difference between context engineering and prompt engineering?

    Prompt engineering is writing and organizing the instructions you give a model. Context engineering is curating everything the model sees at inference time, including those instructions but also tools, retrieved data, message history and memory, and keeping that set right over many turns. Prompt engineering is a part of context engineering, not its rival.

  • What is an LLM context window and how big is it?

    An LLM's context window is the input and output token budget for a request, a form of working memory distinct from training data. Current Claude models such as Opus 5.5, Sonnet 5.5 and Haiku 5.5 have a 1M-token window, per November 2026; some older models have 200k. System prompt, messages, tool definitions, tool results, files, images and the output all count toward it.

  • What is the context engineering skill for Claude, and how do you install it?

    "The context engineering skill" usually means Agent Skills for Context Engineering, an open-source GitHub repository by Muratcan Koylan under the MIT license. It is not one skill but a collection, 18 per November 2026, each a SKILL.md folder on topics like context degradation, compression, memory systems and multi-agent patterns. In Claude Code you install it as a plugin through its marketplace; Cursor, Codex and other hosts can load the same folders.

  • What does Anthropic say about context engineering for AI agents?

    Anthropic's engineering post "Effective context engineering for AI agents" (September 29, 2025) treats context as a finite resource with an attention budget. Its core advice: find the smallest set of high-signal tokens that makes the desired outcome likely. In practice that means system prompts at the right altitude, few and clear tools, a handful of canonical examples, just-in-time retrieval, and for long tasks compaction, structured notes and sub-agents.

All 17 questions →

AI memory

  • Is RAG the same as AI memory?

    No. RAG (retrieval-augmented generation) is a retrieval technique: it finds matching passages in a source, usually existing documents, and puts them into the prompt. Memory is keeping, updating and later recalling what the AI learns, such as preferences, decisions and corrections, and many memory systems use RAG to find those memories again, but RAG alone does not decide what to keep or which fact is out of date.

  • What are the types of memory in AI agents?

    AI agents have short-term memory, the context window and session state of one task, and long-term memory outside the model. Long-term memory is sorted by content (episodic, semantic, procedural), by storage form (instruction files, agent notes, a retrieval index, a memory service) and by scope. For a team, scope matters most: does the memory belong to one user, one project, one agent or everyone?

  • What is A-Mem (Agentic Memory for LLM Agents)?

    A-Mem, short for Agentic Memory for LLM Agents, is a 2025 research paper by Wujiang Xu and colleagues, listed for NeurIPS 2025. It stores each memory as a structured note with keywords, tags and a context description, links it to related notes, and lets new notes update older ones, based on the Zettelkasten method. The code is open source on GitHub.

  • Why does AI forget what you told it earlier?

    A language model only knows what you told it if that text is in its context window, the text it sees while answering. A new chat gets your instructions, project context and a selection from memory, but not your earlier conversations in full. In a long chat, accuracy also drops as the window fills, and older messages can be dropped or summarized.

  • Does OpenClaw have memory, and what does the Mem0 plugin add?

    Yes. OpenClaw keeps memory as plain Markdown files in its workspace, such as MEMORY.md and daily notes, and searches them with memory_search. The Mem0 plugin (@mem0/openclaw-mem0) takes over OpenClaw's memory slot: it extracts durable facts after each turn, recalls relevant ones before the next, and stores them in Mem0's cloud or in a self-hosted open source setup.

  • Mem0 vs Zep vs Supermemory vs Letta: which agent memory tool should I use?

    Pick Mem0 for a memory API that extracts facts per user or agent, Zep when facts change over time and you need to know what was true when, and Supermemory for a usage-priced memory and retrieval API with plugins for coding tools. Pick Letta if you are building the agent itself. All four are developer platforms first, though each now also connects to some existing AI clients.

All 19 questions →

MCP

  • What is an MCP server?

    An MCP server is a program that exposes capabilities to AI applications through the Model Context Protocol (MCP): tools the model can call, resources it can read and prompt templates a user can pick. The AI app (the host) connects to it through an MCP client. A server runs locally on your machine over stdio, or remotely over HTTP, and any MCP-capable app can use it.

  • What is Mem0, and is it open source or a paid platform?

    Mem0 is a memory layer for AI agents and apps: it stores facts from conversations outside the model, so an agent remembers across sessions. It comes in two forms. Mem0 Open Source is an Apache 2.0 library or Docker stack you run on your own infrastructure. Mem0 Platform is the hosted version, which as of November 2026 has a Free plan with usage limits and a Custom plan on request.

  • What is the AWS MCP Server and how do you set it up safely?

    The AWS MCP Server is a managed, remote MCP server run by AWS, generally available since May 6, 2026. It gives AI agents such as Claude Code, Cursor, Codex and Kiro a small set of tools: call any of 15,000+ AWS API operations, search and read AWS documentation, and run short Python scripts in a sandbox. It acts with your IAM permissions and costs nothing extra.

  • What is the GitHub MCP server and how should a team set it up?

    The GitHub MCP server is GitHub's own MCP server. It lets an AI tool such as Claude, Copilot, Cursor or Codex work with your repositories, issues, pull requests and Actions. A team usually connects the remote server at https://api.githubcopilot.com/mcp/ with OAuth, enables only the toolsets it needs, and uses read-only mode where the AI should not write.

  • What does the Figma MCP server give a coding agent, and which seat do you need?

    Figma's MCP server gives coding agents such as Claude Code, Cursor, Codex and VS Code a frame's design context: variables, components, layout and Code Connect mappings. The remote server at https://mcp.figma.com/mcp is recommended, on all plans. As of November 2026, read calls are limited per seat: 20 a month for View and Collab seats, and 200 to 600 a day for Dev and Full seats on paid plans.

  • How do you build a custom MCP server, for example in Python?

    Use an official MCP SDK: Python, TypeScript, C#, Go, Rust and Ruby are Tier 1. In Python you install mcp[cli], create an MCPServer, turn functions into tools with @mcp.tool(), and run it over stdio for one machine or Streamable HTTP for a team. Test it with the MCP Inspector, and add token verification before anyone else connects to it.

All 28 questions →

Alternatives and comparisons

  • Which AI knowledge base software fits a team that already uses ChatGPT or Claude?

    As of November 2026, the main AI knowledge base tools for teams are Notion (Business €19.50 per member a month includes Notion Agent and Enterprise Search), Confluence (Rovo credits included in every paid plan), Slite (from $10 per user a month yearly, with Ask and MCP access) and Guru (custom pricing). If your team works in ChatGPT, Claude or Cursor, ask whether those tools reach the knowledge base over MCP.

  • Claude Projects vs ChatGPT Projects and custom GPTs: what is the difference?

    A Claude Project is a workspace with its own chat history, knowledge files and instructions, shareable with colleagues on Team and Enterprise. ChatGPT Projects is OpenAI's equivalent folder of chats, files and instructions. A custom GPT is a reusable assistant with fixed instructions and knowledge, which OpenAI is moving to plugins; all three keep context for one tool and do not move with you.

  • Mem0 vs Graphiti vs Cognee vs Honcho: which open source agent memory fits?

    All four are open source memory layers for AI agents, built around different ideas. Mem0 extracts facts from conversations and stores them per user or agent. Graphiti, the engine inside Zep, builds a temporal context graph that keeps track of when facts were true. Cognee turns documents, code and conversations into a knowledge graph. Honcho models the people and agents in a conversation. Pick by what you need to remember.

  • ChatGPT Business vs Claude Team: which should your team choose?

    As of November 2026, ChatGPT Business and Claude Team both cost $20 per user a month billed annually and $25 billed monthly, starting at two users. Both offer a pricier seat for heavy users. ChatGPT Business includes Codex; Claude Team includes Claude Code and shared Projects, for 2 to 150 seats.

  • What are Glean alternatives for a small team?

    Glean publishes no price and sells through a demo; it is built for large organizations. As of November 2026, the alternatives for a small team are Dust (from €24 per seat a month, for teams up to 100 people), Microsoft 365 Copilot if your work lives in Microsoft, and ChatGPT Business or Claude Team with their connectors. The right one depends on where your knowledge sits today.

  • What are the best NotebookLM alternatives for teams?

    NotebookLM, renamed Gemini Notebook in July 2026, is strong at answering questions from your own sources. For a team, the alternatives are shared Projects on Claude Team, the ChatGPT Business workspace, or Dust for agents over company knowledge. Teams staying with Google should first check what their Workspace plan includes; Gemini Notebook Enterprise adds admin and security controls.

Team knowledge and AI

  • How do you set up an AI chatbot on your company knowledge base?

    Start with who will ask the questions. For colleagues, the quickest route is a shared Claude project, a custom GPT or a Copilot Studio agent with your documents as knowledge. For customers, a support AI agent such as Zendesk AI agents or Intercom Fin answers from your help center. Build your own RAG chatbot only when neither fits your data, channel or control needs.

  • What is a knowledge base in AI?

    A knowledge base in AI is the store of information a system reasons or answers from. In classic AI, such as the expert systems of the 1970s and 1980s, it held facts and rules that an inference engine applied. In today's AI tools it usually means a set of documents, notes or records that a language model searches before it answers, a method called retrieval-augmented generation (RAG).

  • Claude Projects or NotebookLM for a team's documents?

    Pick Claude Projects when the team wants to work with its documents, writing and deciding with fixed instructions, inside one organization on a Team or Enterprise plan. Pick NotebookLM, which Google's help now calls Gemini Notebook, when the team wants to study a set of sources, with clear limits per notebook and sharing by email or link. Neither one shares what members learn in their own chats.

  • How do you build a knowledge base for AI agents?

    A knowledge base for AI agents holds what an agent cannot find or guess on its own: the team's rules, the decisions behind them, the words it uses, and which source is leading per topic. Keep the always-loaded part short, split the rest into one topic per file that loads only when relevant, and give every entry an owner and a date so outdated knowledge gets removed.

  • What is Karpathy's LLM Wiki, and does it work as a team knowledge base?

    LLM Wiki is an idea Andrej Karpathy published as a GitHub Gist on 4 April 2026. Instead of searching raw documents on every question, an LLM agent reads each new source once and updates a persistent, interlinked Markdown wiki, which you browse in Obsidian. It works well for one person's research. For a team, questions of shared editing, permissions and many tools remain open.

  • How do you make company documents searchable for AI?

    There are three ways. You upload documents into an AI project or notebook (fast, but a copy that goes stale), you connect the AI to the source system, with freshness depending on the connector and its update schedule, or you add a layer underneath that tracks every source and serves every AI tool. Which fits depends on how often your documents change and how many tools you use.

All 9 questions →

By type of team

  • How does a support team get its AI to use what agents already solved in past tickets?

    Not by pointing the AI at old tickets. In Zendesk, AI agents answer from connected knowledge sources: your help center and external content. What an agent solved in a ticket must first be made available through a connected knowledge source, such as a help center article or external knowledge base. Zendesk's Knowledge copilot can draft articles from ticket data; someone still reviews and publishes them.

  • How does an accounting firm share AI memory without mixing up client files?

    Organize memory per client file, not per employee. In Claude that means one project per client on a Team or Enterprise plan, shared with the people on the file, with Use account memory turned off in that project. ChatGPT Business shared projects use project-only memory automatically. Personal AI memory is the risk: it follows a staff member into every chat, so client A can surface in work for client B.

  • Can a law firm share AI memory across matters and lawyers?

    Firm knowledge without client information, like workflows, templates and researched legal questions, can be shared. Client information belongs to the matter. In the US, ABA Formal Opinion 512 (July 29, 2024) says that for self-learning AI tools that can disclose it, informed client consent is needed first, even if only lawyers at the same firm use the tool.

  • How can a consulting firm reuse what it learned on past engagements with AI, without mixing client data?

    Keep two layers apart. Firm know-how (methods, templates, anonymized lessons) goes into a project shared with the whole firm. Client material goes into one project per engagement, shared only with that team; in Claude you also turn off Use account memory there, and in ChatGPT shared projects use project-only memory by default. Personal AI memory is the risk: it carries one client into every other chat.

  • Can a healthcare team share AI projects and memory when patient data is involved?

    If your team is subject to HIPAA, use a service covered by a Business Associate Agreement (BAA) for protected health information, and only covered features. OpenAI offers no BAA for ChatGPT Business, and Anthropic says Claude Team plans cannot enable HIPAA; both require an Enterprise-level setup. Within an eligible workspace, check feature by feature: OpenAI lists projects and shared memory as covered, Anthropic points to its Implementation Guide.

  • Can a marketing agency share Claude Projects and AI memory across its team, client by client?

    Projects yes, memory no. On Claude Team or Enterprise you can share one project per client with Can view or Can edit, and its instructions and knowledge are shared. Claude's memory stays per employee and owners can't view it, so a client's tone of voice belongs in project instructions, with Use account memory turned off so client A's voice does not leak into client B.

All 8 questions →

The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.