How to build an agent

On the Create page, describe your agent in a single sentence — e.g. “An agent that turns a product description into 5 punchy ad headlines.” AI Genesis drafts a full configuration: a name, tagline, description, the input fields it needs, a suggested price, and a starter system prompt.

You land in the reviewscreen — a live editor with a preview of the agent's page beside it. Everything the generator wrote is yours to change before publishing. The sections below walk through each control you'll find there (and on the Edit screen later).

The name, tagline, and descriptionare what visitors see on the agent's page and across the marketplace, so make them concrete about what the agent does and for whom.

Every agent gets its own subdomain (e.g. your-agent.aigenesis.fun) — a real, linkable, embeddable product page. The slug is editable while you create and locked once published, so the link you share stays stable.

Choose the model that runs the agent. Stronger tiers reason better on harder tasks but cost more compute, so each charged run adds a small flat surcharge (charged silently from credits, 100% to the platform — your 90% of the base price is untouched).

Turn on only what the agent needs. Each capability adds a small, clearly-shown surcharge when it's actually used:

Give the agent source material so it answers from your facts, not just its training. Paste text or add URLs (up to 20 sources) and the relevant material is fed into the model at run time. Your sources are private — stored where the public can never read them.

Connect your own HTTPS endpoints (up to 6) so the agent can look things up or trigger actions mid-answer — check inventory, create a ticket, fetch a price. The model decides when to call a tool, uses the live result, and continues. Each tool call adds a small per-call surcharge.

A workflow turns a chat agent into a pipeline that produces one finished deliverable. You define an ordered list of up to 5 steps, each with a short name and an instruction. At run time the steps execute in order — every step sees the original task plus everything the earlier steps produced — and the output of the final step is what the user receives. Leave the workflow empty for a normal chat agent.

Here's a three-step “Market Brief” workflow:

Run with the task “State of the EV battery market in 2026”, the steps chain together:

• Research sees only the task and returns three source summaries.
• Analyze sees the task plus the research, and returns the five themes.
• Write brief sees the task plus both prior outputs, and writes the final summary — the only thing the user sees.

A good workflow makes each step's output genuinely useful input to the next, and phrases the final step as the deliverable you want delivered. Steps beyond the first add a small +1 credit surcharge each (100% platform). Your step instructions are private, like your system prompt.

Switch on “Remember each user”to keep a private, per-user memory across sessions, so the agent personalizes over time. It's off by default and scoped to each individual user — never shared between them.

Add structured fields (text, long text, or number) that appear on the agent's page, so users hand over exactly what the agent needs to do its job — a product description, a target audience, a word count. The generator suggests a starting set; add, remove, or relabel them freely.

Set the agent's base price in credits. Agents are conversational, so one charge covers a block of 10 messages — users chat freely within the block and are only asked to continue when they start the next one. Model, capability, tool, and extra-workflow-step surcharges are added on top and shown right on the page.

Under Advanced you can hand-edit the system prompt— the agent's core instructions — and the max output length. The generator writes a solid prompt to start; tighten it here to control tone, format, and guardrails. The system prompt is private to you.

Publishing puts the agent live on its own page and the leaderboard (you can pause it anytime from your dashboard). When someone spends credits, you keep 90% of the base price, paid out in real money.

Agents are shareable for profit out of the gate: anyone who shares or embeds yours splits your 90% evenly with you on the runs they bring (45%each) — the platform's cut never changes. See Economics for the full money detail.