Deploy as LLM

Quick Steps (TL;DR)

1. Deploy the agent → Settings → Deployments
2. Toggle LLM ON
3. Go to Test → LLM → Keys, create API key
4. Call /chat/completions with model: "agentId@dev|@prod|@1.0"

When to Use LLM Deployment in SmythOS?

LLM Deployment is ideal when you want to expose your SmythOS agent as a flexible, OpenAI-compatible API. It allows any app, script, or tool that works with GPT to work easily with your own agent logic.

Use it when you need to:

• Integrate with your backend
  Call your agent from Python, Node.js, or other server-side environments.

• Power internal tools or automations
  Use the OpenAI-compatible API in low-code tools like Make, Zapier, or n8n.

• Replace OpenAI with your own logic
  Substitute GPT with a version-controlled agent tailored to your domain.

• Expose a stable API to others Share access with devs or third-party tools using scoped API keys and versioned models.

Key Terms for LLM Deployment

Before starting, here’s a breakdown of key terms you’ll see throughout the guide:

Term	Where you see it	What it means	Example
Agent name	Studio header / list	Friendly display name. Not used in API.	Northern Lights SEO Blogger
Agent ID	In URLs & model selector	Unique ID for your agent; used in API calls.	cmb5b4js00fmjh6333d0foxjr
Version tag	Model dropdown	Which version of the agent to run.	@dev, @prod, @1.0
Model string	Request body	agentId + version tag.	cmb5b4js00fmjh6333d0foxjr@prod
baseURL	Test → LLM → Code	Root OpenAI‑compatible endpoint hosted by SmythOS.	https://llm.emb-stg.smyth.ai/_openai/v1

Copy from the UI

Environments vary. Always copy the baseURL and model directly from Test → LLM for the agent you’re using.

Prerequisites

Before you begin, make sure you:

• Have built and saved an agent in SmythOS Studio
• Are logged into your SmythOS account
• Know which environment you want to deploy to (@dev, @prod, or versioned)

Step 1: Toggle On the LLM Deployment

1. Open your agent in Studio.
2. Go to Settings → Deployments.
3. Toggle LLM to On.

This exposes your agent behind an OpenAI‑compatible interface.

Step 2: Copy Your Agent’s baseURL

Click Test (top‑right) → LLM tab → Code.
Copy the Base URL, for example:

https://llm.emb-stg.smyth.ai/_openai/v1

What is baseURL?
The root of the OpenAI‑style API that SmythOS hosts for your agent. You’ll call standard paths like /chat/completions against this URL.

Step 3: Select a Model Version to Use

From Test → LLM use the Version dropdown or copy the string shown in example requests:

• agentId@dev — current in‑development agent
• agentId@prod — latest deployed agent
• agentId@x.y — specific deployed version (e.g., @1.0)

Your final model string looks like: cmb5b4js00fmjh6333d0foxjr@prod.

Step 4: Generate an AgentLLM API Key

In Test → LLM → Keys:

1. Click Create new key.
2. Name it (e.g., “Postman”, “Backend”).
3. Copy the generated key (sk‑…) and store it securely (env var recommended).

Security

Treat AgentLLM keys like any production secret. Do not embed them in client‑side code.

Step 5: Send Your First Request

All examples call the OpenAI‑compatible /chat/completions using your baseURL, model, and AgentLLM key.

Node.js Example

import OpenAI from "openai";

const openai = new OpenAI({
  apiKey: process.env.SMYTHOS_AGENTLLM_KEY,
  baseURL: "https://llm.emb-stg.smyth.ai/_openai/v1",

});

const response = await openai.chat.completions.create({
  model: "cmb5b4js00fmjh6333d0foxjr@dev",
  // Available:
  // ...@dev  -> current in-development agent
  // ...@prod -> latest deployed agent
  // ...@1.0  -> specific deployed version
  messages: [{ role: "user", content: "Hello, what can you do?" }],
  stream: false,
});

console.log(response?.choices);

Python Example

from openai import OpenAI
import os

client = OpenAI(
  api_key=os.getenv("SMYTHOS_AGENTLLM_KEY"),
  base_url="https://llm.emb-stg.smyth.ai/_openai/v1"
)

response = client.chat.completions.create(
  model="cmb5b4js00fmjh6333d0foxjr@dev",
  messages=[{"role": "user", "content": "Hello, what can you do?"}],
)

print(response.choices)

cURL Example

Validate Your JSON Requests

You can use JSONLint or Postman’s visual editor to check the structure of your request body before sending.

# Using a literal key
curl https://llm.emb-stg.smyth.ai/_openai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-a937f3a5-3aee-487b-bc2c-432a287aee65" \
  -d '{
  "model": "cmb5b4js00fmjh6333d0foxjr@dev",
  "messages": [{"role":"user","content":"Hello, what can you do?"}]
}'

# Using an env var
curl https://llm.emb-stg.smyth.ai/_openai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SMYTHOS_AGENTLLM_KEY" \
  -d '{
  "model": "cmb5b4js00fmjh6333d0foxjr@dev",
  "messages": [{"role":"user","content":"Hello, what can you do?"}]
}'

Raw JSON Payload

{
  "model": "cmb5b4js00fmjh6333d0foxjr@dev",
  "messages": [
    { "role": "user", "content": "Hello, what can you do?" }
  ]
}

Import into Postman Instantly

Paste the cURL example into Postman → Import → Raw text and hit enter. Your request is ready to test.

Test the LLM API Endpoint

This is the simplest valid curl request you can send to your deployed agent.
Use it with Postman, cURL, or any OpenAI-compatible client. Replace "agentId@prod" with your actual model string (like cmb5b4js00fmjh6333d0foxjr@prod):

curl https://YOUR_BASE_URL/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_AGENTLLM_KEY" \
  -d '{
    "model": "YOUR_AGENT_ID@prod",
    "messages": [{ "role": "user", "content": "What can you do?" }]
  }'

Enable Streaming Responses

Set "stream": true in your request to receive responses in real time.
This works just like OpenAI's streaming and is useful for chat UIs or progressive output.

{
  "model": "agentId@prod",
  "stream": true,
  "messages": [
    { "role": "user", "content": "Tell me a story." }
  ]
}

Switching Model Versions (dev, prod, or versioned)

• @dev — fastest iteration; reflects current draft.
• @prod — latest deployed production version.
• @x.y — immutable deployed version (e.g., @1.0).

Switch versions by changing the model string without code changes elsewhere.

Where to copy from

Both baseURL and model appear in Test → LLM (Code & Version dropdown). Copy them to avoid typos and environment mix‑ups.

After You’re Live

Success

Your SmythOS LLM agent is live!

• Exercise the endpoint with cURL/Postman, then integrate via any OpenAI SDK.
• Monitor requests in Settings → Logs.
• Roll out new behavior by deploying a version and updating the @tag.
• Pair this API with a UI, or also Deploy as Chatbot / Deploy as ChatGPT.

You now have a versioned, OpenAI‑compatible API powered by your SmythOS agent.

What's Next?

• Build workflows in SmythOS
• Use chatbot with ChatGPT
• Expose your agent as an API

Troubleshooting

Symptom	Fix
401 Unauthorized	Check Authorization: Bearer {AgentLLM key}. Regenerate under Test → LLM → Keys.
Wrong version responding	Verify the model tag (@dev, @prod, or @x.y).
Network/URL errors	Use the exact baseURL shown in Test → LLM for your environment.
Empty/Unexpected output	Inspect Deployments → Logs and confirm LLM is toggled on.