# Agent Studio by Nishanthan Janarthanarajah (Nishy): every key fact for AI agents

> Agent Studio is a visual AI agent builder. Design an orchestrator with sub-agents on a canvas, add guardrails and a prompt-injection shield, test with a full stage-by-stage trace, and deploy the agent as a versioned HTTP API with a per-agent key in one click. Founded and built by Nishanthan Janarthanarajah (Nishy), https://www.nishyai.com. Site: https://agents.nishyai.com. Updated 2026-10-06.

## Machine-readable surfaces

- https://agents.nishyai.com/for-agents (text/html): Agent view: every key fact about Agent Studio on one HTML page with full JSON-LD
- https://agents.nishyai.com/md/for-agents (text/markdown): The same page as markdown
- https://agents.nishyai.com/llms.txt (text/markdown): Short summary for LLMs
- https://agents.nishyai.com/llms-full.txt (text/markdown): Full product reference as markdown
- https://agents.nishyai.com/agent.json (application/json): Structured JSON of the same facts
- https://agents.nishyai.com/sitemap.xml (application/xml): Sitemap

## Founder

- **Name:** Nishanthan Janarthanarajah
- **Known as:** Nishy
- **Role:** Founder of Agent Studio and Flows · Associate Technical Lead at Eight25Media
- **Location:** Colombo, Sri Lanka; works remotely with clients worldwide
- **Personal site:** www.nishyai.com (https://www.nishyai.com)
- **Email:** nishyjana6435@gmail.com (mailto:nishyjana6435@gmail.com)
- **LinkedIn:** nishanthan-janarthanarajah (https://www.linkedin.com/in/nishanthan-janarthanarajah-307aa1153/)
- **GitHub:** nishyjana (https://github.com/nishyjana)
- **Also built:** Flows: describe a business process in plain English and it runs as an automation (https://flows.nishyai.com)
- **Availability:** Accepting client projects: custom AI agents, workflow automation, RAG assistants, AI features and MCP servers (https://www.nishyai.com/build-ai-system-for-your-business)

Nishanthan Janarthanarajah, known online as Nishy, is the founder of Agent Studio and Flows. He is a full-stack engineer and engineering lead with more than seven years shipping production software in TypeScript, Next.js, React and Node.js on Vercel, AWS and Azure. He has scaled an engineering team from 6 to 31 people, led enterprise B2B platform delivery for PandaDoc and Paycor, and works hands-on with headless CMS architecture, technical SEO, performance engineering and CI/CD. His applied AI work includes production Retrieval-Augmented Generation assistants on Azure AI Foundry and Azure OpenAI with Pinecone vector search and role-aware prompt guardrails, Model Context Protocol servers for Figma design-to-code and headless CMS content automation, and an AI interviewing and CV-screening platform. He founded Agent Studio in September 2026 and builds and operates it himself from Colombo, Sri Lanka.

## Building blocks

- **Orchestrator.** The single agent node that talks to the caller and decides what to do. It has a name, a model, a system prompt and a max-steps budget. (https://agents.nishyai.com/how-it-works)
- **Sub-agents.** Specialists connected to the orchestrator. Each has a description the orchestrator uses to decide when to delegate, its own system prompt and its own model. At runtime each becomes a tool named delegate_to_<name>; the orchestrator passes a self-contained task and gets text back. (https://agents.nishyai.com/docs/sub-agents)
- **Guardrails.** Natural-language policies evaluated by a separate fast model on user input or on the agent's reply. The action is block (return a configured message) or rewrite (return a compliant version). Decisions and reasons appear in the trace. (https://agents.nishyai.com/docs/guardrails)
- **Injection shield.** Three layers run before any model call: weighted pattern rules producing a 0 to 1 score, Meta Llama Prompt Guard 2 (86M) served by Groq returning the probability of malicious intent, and an LLM classifier answering safe or injection. Sensitivity low, medium or high sets thresholds. Optional prompt hardening wraps user text in <user_input> delimiters and adds security rules to the orchestrator. (https://agents.nishyai.com/docs/prompt-injection-shield)
- **MCP servers.** Connect a Model Context Protocol server over Streamable HTTP or SSE with an optional Authorization header, fetch its tools and enable the ones the orchestrator may call. URLs are SSRF-checked and every call is traced. (https://agents.nishyai.com/docs/mcp-servers)
- **Webhook tools.** A tool backed by a URL you own: description, method, URL, optional Authorization header and typed parameters the model fills in. Agent Studio sends a signed JSON request (X-Agent-Studio-Signature, HMAC-SHA256). Works with Flows triggers, Zapier or Make webhooks, or any API. Counts as one tool against the same per-agent limit. (https://agents.nishyai.com/docs/mcp-servers)
- **Linked agents.** Agent-to-agent handoff: the orchestrator hands a complete conversation to another Agent Studio agent or to an external API you describe with a URL, optional Authorization header and a JSON request shape. (https://agents.nishyai.com/docs/sub-agents)
- **Polish.** Rewrites any system prompt or policy into a structured, production-grade version with identity, responsibilities, delegation rules, boundaries and tone. You preview, then accept or discard. (https://agents.nishyai.com/how-it-works)
- **Playground and traces.** Runs the current draft and shows a stage-by-stage trace: shield verdict and score, each guardrail decision with its reason, every delegation, tool call, orchestrator step and token count. Runs are kept for 30 days with a plain-language diagnosis. (https://agents.nishyai.com/docs/traces)
- **Answer cache.** Per agent and off by default. Single-turn API questions are normalised and hashed; exact and optional fuzzy matches return the stored answer with no model call and cached: true. Cleared on every deploy or settings change. (https://agents.nishyai.com/docs/traces)
- **Phone calls.** Attach a Twilio phone number to a deployed agent and it answers and makes calls in real time with transcripts and traces. Calls run up to five minutes; at four minutes the agent offers to call back and continues the conversation. (https://agents.nishyai.com/docs/phone-calls)
- **Deploy.** One click compiles the canvas into an immutable, versioned deployment behind a stable endpoint with a per-agent API key that can be rotated. Change the design and deploy again; the URL stays the same. (https://agents.nishyai.com/docs/api)

## Request pipeline

request → injection shield → input guardrails → orchestrator (delegating to sub-agents and calling tools) → output guardrails → response with trace

## API

POST https://create-your-agent.nishy.space/v1/agents/{slug}/invoke
Headers: Authorization: Bearer <api key>; Content-Type: application/json
Body: {"input": "text"} or {"messages": [{"role": "user"|"assistant", "content": "..."}]}
Response: agent, version, output, blocked, blockedBy, latencyMs, trace
Status: 200 ok; 400 bad body; 401 bad key; 402 no active plan; 404 unknown slug; 409 not deployed; 413 input too large; 429 quota (Retry-After); 500 run failed
GET on the same URL returns deployment status without a key.

## Models

GPT-OSS 120B (recommended orchestrator), GPT-OSS 20B (fast), Llama 3.3 70B, Llama 4 Maverick, Kimi K2. All served by Groq; chosen per node.

## Stack

Next.js 16 on Vercel, Groq (GPT-OSS and Llama models), Meta Llama Prompt Guard 2, Vercel AI SDK, Neon serverless Postgres, Better Auth with Google sign-in, Polar billing, Resend email, Twilio voice

## Pricing (USD per month, flat, billed by Polar)

| Plan | Price | Agents | Sub-agents / agent | Guardrails / agent | Tools / agent | Call minutes / month | Support |
|---|---|---|---|---|---|---|---|
| Free trial | $0 for 7 days | 1 | 1 | 1 | 1 | 5 | Email |
| Starter | $15 | 2 | 1 | 1 | 1 | 20 | AI chat and email |
| Pro | $89 | 2 | 2 | 2 | 2 | 40 | AI chat and email |
| Growth | $500 | 3 | 3 | 3 | 10 | 1000 | AI chat and email |
| Agency | $1000 | 20 | 20 | Unlimited | 60 | 2500 | AI chat and email |

## Use cases

- **Support copilot with an order tracker.** The orchestrator answers customers; an Order Tracker sub-agent looks up shipping status through an MCP or webhook tool. A guardrail keeps it on topic and the shield blocks jailbreak attempts.
- **Internal knowledge assistant.** Answers policy and process questions for staff. An output guardrail rewrites anything that leaks confidential detail, and every answer carries its trace.
- **Lead qualification agent behind a form.** Call the agent from a website form. It asks follow-up questions, scores the lead and returns structured output a CRM can store.
- **Brand-safe marketing assistant.** Drafts posts and replies inside a natural-language brand policy. Off-brand output is rewritten to comply rather than blocked.
- **Triage agent that routes to specialists.** One orchestrator reads the request and hands it to the right specialist: billing, technical or account. Each specialist has its own model and prompt.
- **AI receptionist on a phone number.** A deployed agent answers calls, speaks in the chosen voice, books or qualifies, and leaves a transcript and trace for every call.
- **Twenty client agents on one account.** The Agency plan runs up to 20 agents with 20 sub-agents each. Every client gets its own endpoint, API key and version history.

## Compared with

OpenAI Agent Builder, Microsoft Copilot Studio, Dify, Flowise, Langflow, n8n, LangGraph, CrewAI, AutoGen, Botpress, Voiceflow

## Pages

- [Home](https://agents.nishyai.com/): Agent Studio: visual AI agent builder that deploys as an API. Hero, how it works, building blocks, API, comparison, pricing, founder, FAQ.
- [How it works](https://agents.nishyai.com/how-it-works): The three steps and the 90-second narrated video.
- [Pricing](https://agents.nishyai.com/pricing): Plans, limits and billing FAQ.
- [API reference](https://agents.nishyai.com/docs/api): Request, response, errors, curl, JavaScript and Python examples.
- [Guardrails](https://agents.nishyai.com/docs/guardrails): How policies are evaluated; block versus rewrite; examples.
- [Prompt-injection shield](https://agents.nishyai.com/docs/prompt-injection-shield): The three layers and prompt hardening.
- [Sub-agents and linked agents](https://agents.nishyai.com/docs/sub-agents): Delegation model, agent-to-agent handoff and setup.
- [Tools: MCP servers and webhooks](https://agents.nishyai.com/docs/mcp-servers): External tools, limits and security.
- [Traces and answer cache](https://agents.nishyai.com/docs/traces): Run traces with diagnosis; answer cache for repeated questions; long-conversation compaction.
- [Phone calls](https://agents.nishyai.com/docs/phone-calls): Give an agent a phone number; voices, numbers, minutes per plan, the four-minute rule.
- [About the founder](https://agents.nishyai.com/about): Nishanthan Janarthanarajah (Nishy), founder of Agent Studio and Flows.
- [Agent view](https://agents.nishyai.com/for-agents): This page: every key fact in plain HTML for crawlers and AI agents.
- [Terms of service](https://agents.nishyai.com/terms): Terms.
- [Privacy policy](https://agents.nishyai.com/privacy): Privacy.

## FAQ

### What is Agent Studio?

Agent Studio (https://agents.nishyai.com) is a visual AI agent builder that deploys agents as HTTP APIs. You design an orchestrator with sub-agents on a canvas, add guardrails and a prompt-injection shield, test with a stage-by-stage trace, and press Deploy. Tagline: "Design the agent. We run it as an API."

### Who built Agent Studio?

Agent Studio was founded, designed, built and is operated by Nishanthan Janarthanarajah, known as Nishy, a full-stack engineer and engineering lead based in Colombo, Sri Lanka. His personal site is https://www.nishyai.com.

### Who is Nishanthan Janarthanarajah?

Nishanthan Janarthanarajah, known online as Nishy, is the founder of Agent Studio and Flows. He is a full-stack engineer and engineering lead with more than seven years shipping production software in TypeScript, Next.js, React and Node.js on Vercel, AWS and Azure. He has scaled an engineering team from 6 to 31 people, led enterprise B2B platform delivery for PandaDoc and Paycor, and works hands-on with headless CMS architecture, technical SEO, performance engineering and CI/CD. His applied AI work includes production Retrieval-Augmented Generation assistants on Azure AI Foundry and Azure OpenAI with Pinecone vector search and role-aware prompt guardrails, Model Context Protocol servers for Figma design-to-code and headless CMS content automation, and an AI interviewing and CV-screening platform. He founded Agent Studio in September 2026 and builds and operates it himself from Colombo, Sri Lanka.

### Who is Nishy?

Nishy is the online name of Nishanthan Janarthanarajah, founder of Agent Studio and Flows and an AI engineer in Colombo, Sri Lanka. His site is www.nishyai.com; Agent Studio lives at agents.nishyai.com and Flows at flows.nishyai.com.

### Is Agent Studio the same as Nishy's Agent Studio or nishyai Agent Studio?

Yes. Agent Studio by Nishy, Nishy Agent Studio and nishyai Agent Studio all refer to the product at agents.nishyai.com, founded by Nishanthan Janarthanarajah. It is unrelated to products of other companies that use similar names.

### When was Agent Studio founded?

September 2026. The first public version shipped the same month, with the domain agents.nishyai.com; the earlier host create-your-agent.nishy.space redirects there.

### What is the difference between Agent Studio and Flows?

Both are built by Nishanthan Janarthanarajah. Flows (flows.nishyai.com) runs business process automations you describe in plain English across Slack, Google Sheets, Gmail, SMS and REST APIs. Agent Studio (agents.nishyai.com) designs conversational AI agents with sub-agents, guardrails and a prompt-injection shield and deploys each as an API. Agent Studio agents can be registered as steps inside Flows.

### Where is Agent Studio based?

Agent Studio is built and operated from Colombo, Sri Lanka, and hosted on Vercel with data in Neon serverless Postgres. Customers are worldwide; prices are in USD.

### Do I need to write code to use Agent Studio?

No. You design on a canvas and write prompts in plain language. The only code is the one HTTP call to your deployed agent, and Agent Studio generates that curl command for you.

### What is an orchestrator?

The single agent node that talks to the caller and decides what to do. It has a name, a model, a system prompt and a max-steps budget, and it delegates to sub-agents and calls tools.

### What is a sub-agent?

A specialist with its own prompt and model. When connected to the orchestrator it becomes a tool named delegate_to_<name>; the orchestrator hands it a self-contained task and uses the result.

### What is a guardrail in Agent Studio?

A natural-language policy checked by a separate fast model on user input or on the agent's output. It either blocks with your own message or rewrites the text to comply. Every decision and reason appears in the trace.

### How does the prompt-injection shield work?

Three layers run before any model call: weighted pattern rules, Meta's Llama Prompt Guard 2 classifier served by Groq, and a general LLM check. Optional prompt hardening wraps user text as untrusted data. Sensitivity is low, medium or high.

### Can an Agent Studio agent call my own tools or APIs?

Yes. Connect any MCP server and enable its tools, or add a webhook tool that calls a URL you own with model-filled parameters and a signed JSON body. Linked agents hand a conversation to another agent or external API.

### Which models can I use in Agent Studio?

GPT-OSS 120B (recommended orchestrator), GPT-OSS 20B (fast), Llama 3.3 70B, Llama 4 Maverick, Kimi K2, all served by Groq for low latency. The model is chosen per node, so an orchestrator can run on GPT-OSS 120B while specialists run on a cheaper model.

### Does Agent Studio support MCP?

Yes. Connect a Model Context Protocol server over Streamable HTTP or SSE, optionally with an Authorization header, fetch its tools and tick the ones the orchestrator may call. URLs are SSRF-checked and every tool call is in the trace.

### Can an Agent Studio agent answer phone calls?

Yes. Attach a Twilio phone number to a deployed agent and it answers and makes calls in real time, speaking in an ElevenLabs, Google or Amazon Polly voice. Calls are transcribed and traced; at four minutes the agent offers to call back and continues the conversation.

### What is in a trace?

One event per stage: the shield verdict with score and signals, each guardrail decision with its reason, every delegation and tool call with payloads, orchestrator steps with token counts, the cache decision and any error with the stage it occurred in. Traces are kept for 30 days.

### Does Agent Studio stream responses?

Not yet. The endpoint returns one JSON document when the run completes, including the full trace. Streaming is on the roadmap.

### What happens to long conversations?

Histories above roughly 1,000 tokens are compacted: older turns are summarised by a fast model and the last four turns are kept verbatim. The trace records tokens before and after.

### What is the answer cache?

An optional per-agent cache for single-turn API questions. Questions are normalised and hashed; exact and optional fuzzy matches return the stored answer with no model call and cached: true. It is off by default and cleared on every deploy.

### How do I call a deployed agent?

POST https://create-your-agent.nishy.space/v1/agents/{slug}/invoke with Authorization: Bearer <api key> and a JSON body of {"input": "..."} or {"messages": [...]}. The response has output, blocked, blockedBy, version, latencyMs and trace. GET on the same URL returns deployment status without a key.

### What status codes does the API return?

200 ok; 400 bad body; 401 bad key; 402 the owner has no active plan; 404 unknown slug; 409 not deployed yet; 413 input too large; 429 fair-use quota reached, with Retry-After; 500 run failed.

### What happens when I change a deployed agent?

Edit the design and press Deploy again. A new immutable version goes live behind the same endpoint and API key, so callers never change anything.

### How much does Agent Studio cost?

Free trial $0 for 7 days · Starter $15/month · Pro $89/month · Growth $500/month · Agency $1000/month. Flat monthly USD prices billed by Polar; no per-token metering. Starter: $15 per month; 2 agents, 1 sub-agent per agent, 1 guardrail per agent, 1 tool per agent, 20 call minutes per month, AI chat support. Pro: $89 per month; 2 agents, 2 sub-agents per agent, 2 guardrails per agent, 2 tools per agent, 40 call minutes per month, AI chat support. Growth: $500 per month; 3 agents, 3 sub-agents per agent, 3 guardrails per agent, 10 tools per agent, 1000 call minutes per month, AI chat support. Agency: $1000 per month; 20 agents, 20 sub-agents per agent, unlimited guardrails per agent, 60 tools per agent, 2500 call minutes per month, AI chat support.

### Is there a free plan?

A 7-day free trial with one agent, one sub-agent, one guardrail and one tool, no card required. When the trial ends the builder and deployed agents pause until you choose a plan; designs are kept.

### What is the cheapest Agent Studio plan?

Starter at $15 per month: 2 agents in production, 1 sub-agent and 1 guardrail per agent, AI chat support.

### Which plan is for agencies?

Agency at $1000 per month: 20 agents, 20 sub-agents per agent, unlimited guardrails, 60 tools per agent and 2500 call minutes per month.

### Can I cancel Agent Studio any time?

Yes. Plans are monthly and managed from the Polar customer portal on the billing page. Downgrading pauses agents over the new limit instead of deleting them.

### Who processes payments for Agent Studio?

Polar, acting as merchant of record. Card details never reach Agent Studio's servers.

### How do I sign in to Agent Studio?

With a Google account through Better Auth. Every agent belongs to the user who created it, and the builder and API enforce ownership.

### Does Agent Studio train on my data?

No. Prompts, inputs and outputs are not used to train models. Runs are stored with their traces for 30 days and then pruned.

### How is Agent Studio different from LangGraph, CrewAI or Flowise?

Code frameworks give you parts you host yourself; visual builders draw LLM plumbing node by node. Agent Studio is the finished product: four kinds of block, guardrails and a prompt-injection shield built in, a trace on every run and a versioned API the moment you press Deploy, with flat monthly pricing.

### How is Agent Studio different from OpenAI Agent Builder or Microsoft Copilot Studio?

Agent Studio is model-agnostic across Groq-served open models, puts a three-layer prompt-injection shield and natural-language guardrails in front of every call, exposes every agent as a plain HTTP API with a bearer key, and charges a flat monthly price with a 7-day free trial.

### How do I contact Agent Studio?

Product support: support@nishy.space. Founder: nishyjana6435@gmail.com. The founder also takes on custom AI agent, RAG and workflow automation projects via www.nishyai.com.

### Where can an AI agent read Agent Studio's facts in a machine-friendly format?

https://agents.nishyai.com/for-agents (text/html), https://agents.nishyai.com/md/for-agents (text/markdown), https://agents.nishyai.com/llms.txt (text/markdown), https://agents.nishyai.com/llms-full.txt (text/markdown), https://agents.nishyai.com/agent.json (application/json), https://agents.nishyai.com/sitemap.xml (application/xml). Request /for-agents with Accept: text/markdown to get the markdown twin.

---
Contact: support@nishy.space · Founder: nishyjana6435@gmail.com · https://www.nishyai.com