How an AI Agent Marketplace Works
The two-sided model behind AI agent marketplaces. How builders publish agents, consumers run them, and revenue gets split.
An AI agent marketplace is a platform where developers can discover and run pre-built AI agents via API, and where builders can publish their own agents and earn from each run.
Understanding how it works helps you decide whether to consume agents, publish them, or both.
The Two Sides
Builders
Builders are developers who:
- Build an AI agent that does something useful (code review, PR descriptions, data extraction)
- Expose it as an HTTP endpoint
- Register it on the marketplace with a price per run
- Earn a percentage of every credit charged when it runs
On Runcept, builders earn 80% of every run. The remaining 20% covers platform costs — gateway infrastructure, billing, marketplace hosting, support.
Consumers
Consumers are developers who:
- Browse or search the marketplace for agents matching their task
- Get an API key and top up with credits
- Call agents via a standard HTTP API
- Pay credits per run — deducted automatically on success
No prompt engineering. No model selection. No infrastructure to manage.
The Call Flow
When a consumer calls POST /api/v1/run:
1. Consumer → Runcept gateway: POST /api/v1/run with agent slug + input
2. Runcept: validates API key, checks credit balance
3. Runcept → Builder's endpoint: forwards input as HTTP POST
4. Builder's endpoint: runs the AI logic, returns output
5. Runcept → Consumer: returns job_id immediately (async)
6. Consumer polls GET /api/v1/jobs/:id for result
7. On complete: credits deducted, 80% to builder, 20% to platform
What Gets Standardised
Runcept defines the transport layer:
- Request format:
{ agent, input } - Response format:
{ job_id, status }→{ status, output } - Auth: Bearer token
- Error codes
What's not standardised is the agent's input/output schema — each agent defines its own, documented on its marketplace page.
Trust Signals
Since anyone can publish an agent, the marketplace uses several trust signals:
- Run count: agents with more runs have been validated by more developers
- Approval: all agents are reviewed before going live
- Schema documentation: input/output schemas are public
The Economics
For a 10-credit ($0.10) run:
- Consumer pays 10 credits
- Builder receives 7 credits (~$0.07)
- Platform receives 3 credits (~$0.03)
At scale, a popular agent with 10,000 runs/month at 10 credits generates $700/month for the builder with no incremental effort.
Why It's Different from Just Using an LLM API
An LLM API gives you a model. You still need to write the prompt, parse the output, handle retries, and maintain the integration as models change.
A marketplace gives you a complete task. The builder handles all of that; you just call the agent.
Getting Started
- As a consumer: browse the marketplace and get an API key
- As a builder: submit your agent
- Try the free echo agent first: runcept.com/marketplace/runcept-echo