Teach your agent OmniRail once.

What your agent can then do

• Any model, on its own. Picks a model from the live catalog and calls it: Claude, GPT, Gemini, DeepSeek and the rest, for a second opinion or cheap bulk work.
• Images, video and voice. Generates and edits images, makes videos, speaks and transcribes.
• The web and PDFs. Grounds answers with live web results and reads PDFs with any model.
• Search over your data. Embeds and reranks documents for search and RAG.
• Decides before it acts. Asks a decision model a yes/no question about a risky step and gets a probability back.
• Swaps on Robinhood Chain. Quotes, asks you, then signs with its own dedicated wallet. Never yours.
• Within your budget. Asks before anything over $1, picks the cheapest model that does the job and tells you what each call cost.
• Careful with secrets. Never prints your key and never asks for your main wallet.

How it works

1. Your agent only sees the skill's one-paragraph description until a task matches it: generating an image, calling another model, a swap.
2. Then it reads SKILL.md: check the connection, the rules, which tool or endpoint to use for what.
3. Details (parameters, payments, errors) sit in short reference files it opens only when needed, so the skill costs almost no context.
4. It is generated from the same guide as /agents.md, so it always matches what OmniRail offers today. To update, run the install again (the plugin updates with /plugin update).