AMP + Custom GPT Setup
AMP allows most AI models to connect with your system and send and receive code and coding instructions directly. It works with Custom GPTs, Codex, Claude Code, and any AI that can connect over HTTP, through a tunnel or bridge like ngrok.
What it lets you do
- A Custom GPT that reads your project files, searches and traces code, and sends changes Verbatim records, with undo.
- Keep working with ChatGPT from your phone or another computer while Verbatim runs at home.
- Agents on the same computer, like Claude Code or Codex, connect directly with the same key, no tunnel needed.
- Every change arrives as a Recording Sheet: logged, diffed and reversible.
How it fits together
| Part | Job |
|---|---|
| Your Custom GPT | Conversation, reasoning and action requests |
| ngrok | Your permanent private address, forwarding to Verbatim's AMP port 5056 |
| AMP key | The lock: Verbatim refuses any request without it |
| Verbatim | Reads, records, diffs and undoes, on your computer |
Set it up by hand
Install ngrok and add your authtoken
Create a free account at dashboard.ngrok.com, install ngrok (Windows: winget install ngrok.ngrok, Mac: brew install ngrok), then run the command below with the authtoken from your dashboard:
ngrok config add-authtoken YOUR_NGROK_AUTHTOKEN
Start the tunnel
With Verbatim running:
ngrok http 5056
5056 is Verbatim's AMP port. ngrok shows your address, like something.ngrok-free.dev. Free accounts keep the same address, so your GPT is set up once.
Create your AMP key
The Guided Setup's CREATE KEY does this safely. By hand: open C:\Nodarama\nv\settings\user.json (or your install folder's nv\settings\user.json). Right after the very first {, add this line, ending with a comma, and change nothing else:
"bridgeAuth": { "apiKey": "a-long-random-phrase-only-you-know" },
If "bridgeAuth" is already there, change its value instead. Save; no restart is needed.
powershell -c "Get-Content C:\Nodarama\nv\settings\user.json -Raw | ConvertFrom-Json | Out-Null; 'valid'"Create your GPT and paste the instructions
In ChatGPT: Create → Configure. Paste the instructions from Verbatim (Guided Setup → COPY INSTRUCTIONS, or docs/guides/Custom_GPT_Instructions.txt in your install) into the Instructions box.
Add the actions
Create new action → Import from URL, and paste your address with /amp/openapi.yaml on the end. Verbatim serves the schema with your address already filled in:
https://YOUR-ADDRESS.ngrok-free.dev/amp/openapi.yaml
No tunnel running yet? Fill your address into the schema below and paste or upload it instead.
Set the authentication
In the action: Authentication → API Key → Auth Type Custom → Custom Header Name x-amp-token → API Key: your AMP key.
Save privately and test
Save with sharing set to Only me; this GPT can change your project. Then ask it: "Check my Verbatim connection." and choose Allow.
Schema with your address filled in
Type your ngrok address. The schema is filled in right here in your browser; nothing is sent anywhere. Your AMP key never goes in the schema: it belongs only in the GPT's Authentication box.
When something doesn't work
| You see | Fix |
|---|---|
amp_token_unconfigured | Verbatim has no AMP key: step 3 |
connection_authority_required (401) | The GPT's key doesn't match: paste it again (step 6) |
| No answer at all | The tunnel isn't running: step 2 |
| Import from URL fails | Use the filled-in schema above |
Safety notes
- Never paste your AMP key or ngrok authtoken into a chat.
- Keep your GPT on Only me, and don't publish your ngrok address.
- Use review and undo for changes you don't fully trust.
Final check
The goal is not to let ChatGPT freely write to your computer. The goal is to let ChatGPT communicate with Nodarama Verbatim, while Verbatim remains the controlled local execution layer.