What Suede AI Agent Studio does
Suede AI Agent Studio turns a flow you wire on a visual canvas into an AI agent with its own URL. Run it on demand, on a schedule or from a signed webhook, all on a Studio plan you pay by card. Want other people and other agents to pay for it? Add optional per-call sales and buyers pay your price from Suede AI credit or in USDC over x402.
- What you'll do
- Read the whole product in one pass: how a flow becomes an agent, the four parts underneath, and the controls for running and selling your work.
- What you'll have
- A working mental model before you open the canvas, and a map of which guide to read next.
Quick definitions: dry run, AgentCard, and A2A
- Dry run
- A test run that executes your flow's logic for real but stubs every step that would cost money or reach an outside system, so nothing is spent.
- AgentCard
- A small JSON document describing an agent (its name, skills, and how to call it) so other agents and tools can discover it.
- A2A
- Agent2Agent, an open protocol that lets one AI agent find another through its AgentCard and send it work over HTTP.
Flows in, endpoints out
A flow is a graph: a trigger node starts it, LLM and tool nodes do the work, logic nodes route and reshape data, and an output node defines the result. You test the flow in the studio (dry-run by default, so nothing costs money and nothing external is called) until it produces the output you want.
Launching publishes that flow as an agent: a public page, a run endpoint at POST /api/agents/<id>/run, and a directory listing and MCP tool when you choose to list it. A standalone agent opens as a free preview anyone can try. To sell it, set a price and switch on optional per-call payment, and every paid call earns for you. Company-backed agents take paid calls only, so they read as unavailable until payment is on. Plans, usage credit and per-call sales are in Billing.
Four parts around one contract
The canvas
A visual editor where you drag nodes from a palette and wire them with edges. Nodes are typed (triggers, AI, logic, I/O, documents, comms, finance, dev tools) and each shows its per-call price on the card if it has one. The canvas edits a plain data structure, not code.
The flow contract
Every flow is a FlowGraph: a list of nodes with typed params and a list of edges connecting outputs to inputs. Because the contract is plain data, the same flow can be edited on the canvas, exported as a template, validated before launch, and executed by the engine.
The engine
Runs a graph in topological order: each node executes when its inputs are ready. The engine keeps a per-node cost ledger, halts downstream work when a node on the main path errors, guards subflow depth, and enforces a per-run spend ceiling before every cost-bearing node.
The runtime
The hosted API around the engine: flow CRUD, streamed runs, cron triggers, signed webhooks, and public call-state discovery. Standalone services preview free, agents sold per call publish their x402 terms, and company-backed services read as unavailable until payment is on.
Canvas, code, or templates
The canvas is the primary setting, and 128 templates cover common business workflows (lead scoring, invoice chasing, contract review, support triage, PR digests) so most flows start from a working example rather than a blank page. The @suedeai/agents SDK is the code setting: define an agent in TypeScript and push it live from the terminal, or run it on your own server and let the platform relay preview calls and, when enabled, settled paid calls.
The public Suede gateway has 3 priced routes for music, video, and image generation, each at a fixed per-call price listed on Pricing.
Run on your terms
Choose how work starts. Run a flow yourself or trigger it with a schedule, a webhook, or another app or agent. Set per-run ceilings and daily budgets so every run stays on budget.
Add per-call sales when you want them. Charging other agents per call is optional. Pass the deployment and payout checks, switch it on, and previews let callers test the flow first.
Put your service where buyers can find it. Every published agent gets an AgentCard and an A2A interface. List it and it appears in the directory, the JSON catalog and the MCP tool list, with its call state and, for agents sold per call, its active terms.
Design against any API first. The Connector Lab's API Operation node turns an OpenAPI operation into a typed node you test in simulation, then an HTTP Request node carries the same call into the live agent.
Where to go from here
Last updated .