Hire an expert your agent can call.

Most skill marketplaces sell you the skill itself. You download the SKILL.md folder onto your machine, so from then on it can be copied, forwarded, or resold, and the author is paid once and loses control of their know-how. Askpert works the other way around: the skill stays sealed on our servers. You never download it. Your own agent calls it over MCP as a tool, and the instructions, knowledge, and code never cross the wire. The seller keeps their method and earns per call, and every listing clears a 20-check inspection before it can go live.

Yes. An expert runs on your own LLM API key. This is BYOK, bring your own key: you supply the key for the model (OpenAI, Gemini, or OpenRouter), so you control the cost and your conversations stay in your own provider account. Each listing names the keys it needs before you hire. Add them once under Settings, API keys, on the Buyer tab, and every expert reuses them.

Experts run on OpenAI, Gemini, or OpenRouter models, and you bring the LLM API key for whichever one the listing names. Some experts also search the web through Exa or Parallel, but those run on the seller's key, so there is nothing extra for you to add. Set your model keys under Settings, API keys.

Two things, and they stay separate. First, the expert's own per-call price. The seller sets it in credits, and every expert listed today is free at 0 credits. If a seller does price an expert, that price shows on the listing before you commit. Second, your own LLM API tokens: every run uses your BYOK model key at your provider's normal rates, so token spend lands on your account and is never bundled into anyone else's bill. Browsing itself needs no account.

An expert only ever receives what you choose to send it. Each listing declares up front whether it accepts files at all, and when it does, your own harness gathers them locally and passes along only what is needed, so your raw files do not leave your machine on their own. If an expert runs code over data, that code runs in an isolated server sandbox with no internet access unless the listing names specific domains, and it can never reach back into your local machine. Every reply also carries a safety notice and machine-readable risk flags, so your harness can pause anything sensitive.

The model call runs on your own LLM API key, so the conversation lives in your provider account under that provider's terms. The seller never sees your messages, and nothing you send reaches other buyers. Usage tracking stores counts and dates, not content.

Buyers bring the LLM API key that runs a published expert's model. Sellers bring the LLM API keys behind everything they build with: the drafting assistants, Skill Studio, and their own test runs. Askpert's own keys fund exactly two things, the publish inspection and the generated avatar art. So no one's LLM usage is ever bundled into someone else's bill.

You run a frozen copy of what passed inspection, byte for byte. A seller's later edits change nothing for you until they re-inspect and re-publish. If a seller unpublishes, new calls stop; nothing you already received is touched.

Buyers get answers, not contents. Instructions, skills, documents, and code stay server-side; the API returns name, description, and reachable domains for a code tool, nothing more. An anti-extraction guard wraps every expert, and the inspection includes checks that it won't reveal its own instructions.