Orchestrator, sub-agents, guardrails, one Deploy button
Design the agent. We run it as an API.
Drag an orchestrator onto the canvas, give it specialists, wrap it in guardrails and a prompt-injection shield, then hit Deploy. Every change ships as a new version behind the same endpoint.
7-day free trial · One agent · No card required
How it works
Ten minutes from blank canvas to a live endpoint
No orchestration code, no prompt plumbing, no separate moderation service. The builder compiles your design into a runtime and hosts it. Read the full walkthrough.
- 01
Design on the canvas
Add an Orchestrator and connect Sub-agents for the work it should delegate. Write each prompt in plain words, then press Polish to turn it into a production-grade prompt.
- 02
Add protection
Drop in Guardrails with the policies your product needs and an Injection Shield that stops jailbreak attempts before the model sees them. Test it in the playground with a full trace.
- 03
Deploy as an API
One click compiles the design into a versioned deployment. Call it with a bearer key from any language. Change the design, deploy again, same URL.
Under the hood
Watch a request travel through your agent
Every call runs the same pipeline: shield, input guardrails, orchestrator and its sub-agents, output guardrails. The trace on the right is what your API returns.
Hover to pause
Incoming message
Where is my order #A1234?
Trace so far
- Nothing yet…
Building blocks
Everything a production agent needs, as nodes
Talks to the caller, decides what to do, and delegates. Pick the model, set the step budget, write the prompt.
Specialists with their own model and prompt. Connect one to the orchestrator and it becomes a tool the orchestrator can call.
Natural-language policies checked on input or output. Block with your own message, or rewrite the text to comply.
Pattern rules, Meta's Llama Prompt Guard 2, and an LLM classifier. Optional hardening treats user input as data, never instructions.
Turn a one-line idea into a structured system prompt or policy. Preview, then accept or discard.
Connect any MCP server and tick its tools, or add a webhook tool that calls a URL you own. Every call shows in the trace.
Versioned deployments, per-agent API keys with rotation, and a stage-by-stage trace on every run.
The API
One endpoint per agent, stable across versions
Send a message or a whole conversation. Get back the answer, whether a guardrail or the shield blocked it, and the trace that explains why.
- Bearer API key per agent, rotate any time
- Blocked responses carry the stage that blocked them
- Deployed versions are immutable; redeploy to change
- Every call is logged with latency and token usage
curl -X POST https://create-your-agent.nishy.space/v1/agents/support-copilot/invoke \
-H "Authorization: Bearer ak_…" \
-H "Content-Type: application/json" \
-d '{"input": "Where is my order #A1234?"}'
{
"version": 3,
"output": "Order A1234 shipped yesterday via UPS…",
"blocked": false,
"trace": [
{ "stage": "shield", "status": "pass" },
{ "stage": "guardrail", "name": "Stay on topic", "status": "pass" },
{ "stage": "delegate", "subagent": "Order Tracker" },
{ "stage": "orchestrator", "steps": 2 }
]
}Why Agent Studio
Frameworks give you parts. Builders give you plumbing. We give you the agent.
Code frameworks are excellent if you want to own the orchestration and host it yourself. Visual builders draw the LLM plumbing, fifty nodes for one agent. Chat platforms are built around widgets. Agent Studio is the finished product on top of those ideas: four blocks, safety built in, and an API the moment you press Deploy.
| Capability | Code frameworks LangGraph, CrewAI, AutoGen | Visual builders Flowise, Langflow, n8n | Chat platforms Botpress, Voiceflow | Agent Studio This product |
|---|---|---|---|---|
| Design the whole agent visually | — | Yes | Yes | Yes |
| Sub-agents as first-class delegation | Partial | Partial | — | Yes |
| Guardrails built in, on input and output | — | Partial | Partial | Yes |
| Prompt-injection shield, on by default | — | — | — | Yes |
| One-click prompt polishing | — | — | — | Yes |
| Stage-by-stage trace on every run | Partial | Partial | — | Yes |
| Deploy as a versioned API in one click | — | Partial | — | Yes |
| Hosting included, nothing to run | — | Partial | Yes | Yes |
| Flat monthly price, no per-token metering | — | Partial | Partial | Yes |
Based on public documentation as of September 2026. “Partial” means possible with extra code, plugins, or self-hosting.
Safety is not a plugin
Guardrails and the injection shield are blocks on the canvas, not libraries you remember to add. Every decision they make is in the trace.
Delegation is a drawing
Connect a sub-agent and it becomes a tool with its own model and prompt. No routing code, no graph definitions.
Deploy is one click, and versioned
Change the design, redeploy, same URL and key. Every run explains itself with scores, reasons, tokens and latency.
Pricing
Flat monthly plans, in USD
Start with a 7-day free trial of one agent. Pick a plan when you are ready to keep it running. Change or cancel any time. Full pricing details.
Built by the founder
Nishanthan Janarthanarajah (Nishy)
Founder of Agent Studio and Flows. Full-stack engineer and engineering lead with 7+ years shipping production systems, and hands-on builder of RAG assistants, MCP servers and agentic automation. Agent Studio is the product he wanted while putting AI agents into real businesses: safety built in, deploy in one click.
FAQ
Questions people ask before they start
Do I need to write code?+
No. You design on a canvas and write prompts in plain language. The only code is the one curl command you use to call your agent, and we generate that for you.
What is a sub-agent?+
A specialist with its own prompt and model. When you connect it to the orchestrator, the orchestrator can hand it tasks and use the result, the way a lead hands work to a teammate.
How does the injection shield work?+
Three layers: fast pattern rules, Meta's Llama Prompt Guard 2 classifier, and a general LLM check. It can also harden your orchestrator so user text is treated as data. Every decision shows up in the trace.
What happens when my trial ends?+
The builder and your deployed agents pause, and we email you a link to choose a plan. Nothing is deleted; pick a plan and everything resumes as you left it.
Which models can I use?+
GPT-OSS 120B and 20B, Llama 3.3 70B, Llama 4 Maverick, and Kimi K2, all served by Groq for low latency. You can choose a model per node.
Can I cancel any time?+
Yes. Plans are monthly and managed from the customer portal on the billing page. Downgrading pauses agents over the new limit instead of deleting them.