You're already in the middle of something.
Let's get a second opinion on it.
Who do you talk to?
- in your assistant
Start a fresh chat
New conversation, outside any project or folder. Sounds fussy, but it matters: in an existing thread your assistant just rehashes whatever you were last talking about, and projects are walled off in both directions. Fresh chat, everything it carries everywhere.
In ChatGPT: New chat, not inside a Project, and not a Temporary chat.
Open a new ChatGPT chat ↗ Paste this and hit enter
Nothing else. No setup, no context, no explaining yourself. The prompt changes with the assistant you picked.
Prompt · ChatGPT · 10 paragraphs · Paste all of itIf your assistant offers to attach this as a file, choose Paste as text. This is an instruction, not a document.
Two replies: a short list, then a write-up. Plain text, no tables.
Only look at decisions about work, money, business, or logistics. If something in what you know about me is a family, medical, or legal-personal matter, skip it entirely. Do not rank it, describe it, allude to it, or explain that you skipped it. Same for which investment, fund, or security to buy, sell, or hold. If unsure, skip it. Fewer than three is fine. If you hold nothing about me, say so in one line and stop.
You hold three separate things about me: what you remember (your memory summary or saved memories), the past chats you can pull up, and the custom instructions or context you were handed at the start of this chat. Use whichever you can actually reach, and say which one each thing came from, because saved memories go stale and you will not notice on your own. If a thing came from your own earlier suggestion rather than from me, say so. Your strength here is reach across a long history. Your weakness is that you smooth things over and leave out the parts that make me look bad. I need those parts.
Read the rest (7 paragraphs)
Now find the decisions I am actually in the middle of right now, not ones I have closed. Before you rank anything, check whether what you are relying on has since changed or is contradicted elsewhere in what you hold. If so, show both versions and say you are not sure which is current. Do not quietly pick one. Then pick the top three and rank them by how much is riding on them, using only what you can point at: whether it can be undone, whether a clock is running, how many people bear real consequences if it goes wrong, and how many times I have come back to it without settling it.
Show me what you found, up to three, numbered, one line each on what the decision is and one line on why you ranked it there. No names. Then stop and ask me to type 1, 2, or 3. That is your whole first reply. Wait for my number.
When I give you a number, write up that one without checking back. The most important part is the section on what cuts against me: anything in our history that contradicts this decision, or that I am glossing over. Write that section first and write it at full length. Do not shorten it to make room for anything else in this prompt. Then cover: what I have said matters to me, what I have said I will not do, who bears the consequences if it goes wrong, anything similar I tried before and how that went, and any deadline or point of no return.
Do not guess. If you do not know something, say so and list it as an open question instead of filling it in. Tell me plainly how much you actually know about me and how far back it goes.
Then give me three things I could do next on this decision, numbered 1, 2, 3. For each: one line on what it is, one on why someone might choose it, and a line starting "Done when:" that names something observable: a document that exists, a date on a calendar, an answer from a specific role. Not "when I feel ready." Put a warning at the start of anything hard to undo. Do not tell me which to pick and do not rank them. Separately, unnumbered, list anything waiting on someone else or blocked, and which role would unblock it.
Before you finish, read back everything you wrote and remove every person's name, company name, address, account number, case or file number, repository path, and commit identifier. Replace people with their role. Keep dollar amounts exact, because the amounts are what the Review turns on.
End with a block headed FOR TERANODE, six labeled lines. Decision question, as one question. Where I am leaning, in my words as you have them, or "no stated leaning"; do not supply one. What is at stake, one line. Open questions, short. Identifiers left in and why. Model name and version, as best you know it; if you are not sure, write "not sure" rather than guessing. Nothing after the block.
- in your assistant
It comes back with three decisions, or fewer. Pick one.
It goes looking for what you're actually in the middle of, ranks them by what's really at stake, and shows its reasoning. You type 1, 2, or 3.
Then paste the whole thing here
All of it, even the parts that look like filler. Especially the part listing what it doesn't know.
Teranode runs the write-up past separate model vendors, none of them the one that wrote it, and comes back with findings. The findings inform your decision. You make it. They are not investment, legal, or tax advice.
The prompt already tells your assistant to take out names, account numbers, and dollar figures tied to a person. If one slips through, the next screen asks you to remove it before anything runs.
Why three different prompts?
Because these models aren't interchangeable. ChatGPT keeps a running summary of everything you have told it and can pull up old chats, and it tends to smooth over the parts that make you look bad. Claude can go read what you actually said instead of a summary, then hedges about it. Grok is the bluntest of the three and keeps the least, so its write-up is shorter and it will say what it does not know.
Each prompt names what its model does well and where it slips, and asks for the unflattering version instead of the polished one.
About five minutes to get to the button. The Review itself takes a couple more.
Work, money, business, and logistics only. The prompt tells your assistant to leave family, medical, and legal matters out, and not to mention that it did.
If your assistant doesn't know you that well, you might get one decision instead of three, or a shorter write-up. Send it anyway. We'd rather have thin and true than padded.
Doing this on behalf of a client or a family member?
The chat is where this belongs.
Chatbot-native is the priority. Connect Teranode once and the Review comes back inside ChatGPT, Claude, or Grok, in the conversation you are already having, with no pasting. The prompt flow above is the fastest way to try it today; this is where it lives.
Claude
In Claude, open Customize, then Connectors, then the plus button, then Add custom connector. Name it Teranode Review and paste the two Claude fields below. The OAuth Client ID sits under Advanced settings. Claude calls it optional and for our server it is not. Leave it blank and Claude registers its own client instead, which fails. Free plans get one custom connector.
https://teranode.ai/api/mcp/native-salesvOEox4vHdDBhYGFB4BwUaL32LohRbmnyThen: Add, Connect, Continue with Google, Accept. In the chat, say “Run an independent Teranode Review.” and confirm the exact recommendation.
Open Claude connector settings ↗ChatGPT
Desktop browser only. Turn on Developer mode under Settings, then Security and login. Then go to chatgpt.com/plugins and use the plus button to add the connector. Name it Teranode Review and paste the Server URL and OAuth Client ID below. The OAuth Client ID sits under Advanced OAuth settings, which ChatGPT keeps collapsed. Expand it and enter the ID. Skip it and ChatGPT registers its own client instead, which fails.
https://teranode.ai/api/mcp/native-salesYV6gWCQHhFI5p3wYuQTdxWKiYddgxOEvIn the chat, open the plus menu, choose Developer mode, tick Teranode Review, then send.
Run Teranode Review on the decision below.Some ChatGPT plans limit connector write actions; if the Review tool is refused, the paste path above works on every plan.
Open ChatGPT connector settings ↗Grok connector support is coming.
Google sign-in only. Email and password signup is not available for new Teranode connector users.
Firm or IT? The pilot setup, token lifecycle, and system prompt block are at /connect/firm.
Talk to Teranode Support. Need help connecting Teranode to ChatGPT, Claude, or Grok, or understanding what this is? Call 1-888-837-6950 (888-TER-6950). The call is on Grok. Paste the write-up here for a Review. The line is not for reviewing a case. Findings are not advice.
After you connect
- Buying another Review. Reviews are bought one at a time on your Teranode account page, signed in with the same email as your chat connection. Checkout runs on Stripe and shows the current price before you pay. A CRD number or advisor affiliation is not required.
- Need another Review mid-conversation? Buy one on the account page, return to the same chat, and ask again. It is available immediately; nothing about your conversation is lost.
- Finding a completed Review. Findings return in the same conversation, together with a link to the signed disposition record. Every Review you have run is also listed on your account page.
Every completed Review you confirm is kept on your account so you can come back to it. It is private to your account and is the start of Continuity, Teranode's context memory for future Reviews.
What Teranode stores, and why. Your sign-in email (your account identity), your purchase history, and each Review you confirm: the exact material reviewed, the findings, and the signed disposition record, kept so you can retrieve them later. Reviews are scoped to your account; other customers cannot see them. Teranode staff can access stored records for support and operations. Do not include SSNs, account numbers, or other identifiers in a Review; the preparation step refuses material that carries them.
In plain terms. We don't keep your client's name. If a Review runs, we keep that Review on your account so you can open it again. We don't use it to train models, and we don't put it in the public dataset.
A private library for the decisions where the stakes were high.
Every Review you confirm is kept on your account: the exact material reviewed, the findings, and the signed disposition record. In your words, not a summary. Over time it becomes a private library of the decisions that mattered, the ones you would want to find again.
Own Your Own Node.
Persistent memory for one person, built to run six months ahead of what the frontier models ship, and yours to keep. Teranode will offer it as a subscription. Join the list and we will write to you when Own Your Own Node is live. Nothing else, no newsletter.
