The four everyone has heard of, and the three that break the rental model. Read the friendliness column first if you are not technical. Read the last two columns if you care about cost and leverage.
ChatGPT (chat)
OpenAI · cloud
The default. Excellent at writing, explaining and summarising.
FriendlinessThe benchmark. Nothing to install, nothing to explain to a colleague.
CustomisabilityCustom instructions and Custom GPTs. It cannot see your folders or calendar unless you upload or connect them.
UsabilityYou open it and drive it. Nothing happens while you are in a meeting.
Model-agnosticNo. You get OpenAI's models, at OpenAI's pace and prices.
PriceFree tier; Plus US$20/month.
The catchYour work sits in someone else's cloud, and the assistant exists only while you are looking at it.
ChatGPT Work
OpenAI · cloud · business seats
The same assistant, sold to organisations with admin controls and bigger limits.
FriendlinessHigh. If you know ChatGPT, you know this.
CustomisabilityWorkspace administration, connectors, credit top-ups. No access to your local machine.
UsabilityOrganisation-wide chat with central billing. Premium seats get roughly five times the usage and no five-hour limit.
Model-agnosticNo. OpenAI models only.
PriceStandard seat US$25/month, or US$20/month billed annually. Premium seat US$125/month, or US$100 annual. Two seats minimum.
Where it fitsThe right answer if the Centre wants one managed vendor and does not mind renting per seat.
Claude Cowork
Anthropic · desktop + cloud
Hand it a goal, walk away, come back to finished work.
FriendlinessHigh. Anthropic designed it explicitly for people who do not want a terminal.
CustomisabilityIt reads and writes files and folders you point it at, runs scheduled tasks, and connects to outside tools through MCP.
UsabilityBest-in-class for document work: spreadsheets with working formulas, decks, formatted reports. Sessions run in Anthropic's cloud and files live with your Claude account.
Model-agnosticNo. Claude models only.
PricePro US$20/month (US$17 annual). Max from US$100. Team standard seat US$25 (US$20 annual), premium US$125 (US$100 annual). Enterprise US$20/seat plus usage.
The catchPer-seat, single-vendor, and your working files live in their cloud.
Gemini Spark
Google · cloud
A 24/7 agent that keeps working with your laptop shut.
FriendlinessHigh, on one condition: you already work inside Gmail, Calendar and Drive.
CustomisabilityTasks, skills and schedules, wired natively into Google's apps.
UsabilityRuns in the background in Google's cloud. Strong for inbox triage, digests and research.
Model-agnosticNo. Gemini models only.
PriceBundled with Google AI Pro (S$28.99/month in Singapore) or Ultra (from S$359.98/month).
Watch outPersonal Google accounts only for now (not work or school accounts), 18 and over, and not available in some regions including the UK and the EU.
Vellum
Vellum · Mac app or their cloud
A personal assistant with one memory across every device you use.
FriendlinessHigh. Native Mac app, guided first run, no terminal.
CustomisabilitySkills and channels; it is open source, so it can also be self-hosted on your own hardware with your own keys.
UsabilityMac, iPhone, web, voice, email, Telegram, Slack, Teams and Discord all share one memory. Good at "remember this and remind me".
Model-agnosticYes. It passes model costs through at cost rather than marking them up.
PriceFree Base plan with pay-as-you-go credits. Mighty US$30/month, Super US$100/month, Ultra US$200/month (Super and Ultra add a US$10/month platform fee).
Where it fitsThe closest ready-made product to what we are proposing, at three to twenty times the running cost.
OpenClaw
OpenClaw Foundation · your machine
Open source, self-hosted, and answers you in WhatsApp, Slack or Telegram.
FriendlinessLowest in this table. There is no vendor to call: you install it, update it and fix it.
CustomisabilityHighest. MIT licensed, plugin channels for 20+ messaging platforms, native apps, one gateway for a laptop or a team.
UsabilityOnce it runs, it is excellent: the agent lives in the chat apps you already use, on your own infrastructure.
Model-agnosticYes. It is a harness, so the model is whatever you plug in.
PriceSoftware free. What you actually pay is hosting (US$5 to US$24/month for a small server, or nothing if it runs on a Mac you own) plus model usage. Typical single-user spend lands between US$11 and US$35/month light, and US$40 to US$100/month under heavy automation.
Where it fitsThe same category as Hermes Agent. Both are self-hosted harnesses; they differ in the polish of the setup, not the idea.
Hermes Agent
Nous Research · your machine
The one being demonstrated. Open source, model-agnostic, runs on a Mac you already own.
FriendlinessMedium. One installer (or a desktop app), then a setup wizard asks the questions. It is the most technical item in this table, and the guides on this site walk it end to end.
CustomisabilityHighest. Skills, plugins, scheduled jobs, MCP connectors, browser control, file access, and a memory folder you can read and edit yourself.
UsabilityYou talk to it from Telegram, Slack, WhatsApp or its own window. Scheduled jobs run whether or not you are at the desk, and it reports to your phone.
Model-agnosticYes. 300+ models through Nous Portal, or plug in OpenAI, Anthropic, Google, OpenRouter, or any OpenAI-compatible provider.
PriceSoftware free. You pay only for model usage. The real bill on this machine is US$10/month: a US$10 subscription that includes US$70 of model credit, plus web search on a free monthly credit.
Where it fitsFree software, your hardware, your data, and the freedom to change model or provider the day a better one appears.
What actually separates them
Three columns decide this, and none of them is "which one is smartest".
1. Where it runs. Cloud assistants are someone else's computer. Self-hosted ones are yours. For client material, that difference is the whole conversation, and it is the one PDPA cares about.
2. Model-agnostic or not. The four branded assistants tie you to one model family. The three open ones (Vellum, OpenClaw, Hermes Agent) let you swap model or provider without changing your habits. The barrier you are stuck behind is usually this one.
3. Per seat or per machine. Cloud tools charge per person, every month, forever. A self-hosted assistant costs the same whether one advisor uses it or ten, because the bill follows usage, not headcount.