Prompts
Starting points you can copy and paste to your agent — edit the highlighted fields first.
Reach your agent over SSH (inferenceai ssh then hermes chat) or from a connected chat channel. New here? Start with the CLI setup.
Build and host something
A web app on a public URL
Build , serve it on port 3000, and expose it publicly with cloudflared. Give me the URL.
An endpoint that captures every webhook
Stand up an HTTP endpoint on port 3000 that logs every incoming POST — headers and body — to , then expose it with cloudflared and give me the URL.
Reveal a local project on a public URL
I have running on port . Put it on a public URL with cloudflared, confirm it's reachable, and give me the URL.
Work through a folder of files
Pull data out of a folder of PDFs
First, copy your files over: inferenceai cp -r ./invoices vm:/workspace/invoices
I have PDFs in . Extract from each into . Skip any corrupt files and log them to errors.log.
Summarize a stack of documents
First, copy your files over: inferenceai cp -r ./docs vm:/workspace/docs
Read every document in and write a one-paragraph summary of each to , using the filename as a heading.
Organize a messy folder
First, copy your files over: inferenceai cp -r ./dump vm:/workspace/dump
Sort the files in into subfolders by , and write what you moved to moved.txt. Don't delete anything.
Run on a schedule
Generate in bulk
Generate a batch and save it to a file
Generate short, distinct . One per line, no duplicates. Save them to .
Writing your own
These prompts share a few patterns. They hold for anything you ask your agent:
- Describe the outcome, not the steps. Say what you want and let the agent choose the tools and files.
- Give it a way to check its own work. Ask it to run, test, or verify in the same prompt so it iterates instead of stopping.
- Point at a reference. Name a file, folder, or example to match so the result fits what you already have.
- Say where the output goes. Name the file, format, or channel so you end with something you keep.