Version: Next

Agentic LLM

Build applications with ASI:One's agentic models that can autonomously call agents from the Agentverse marketplace and handle complex workflows. These models can discover, coordinate, and execute tasks through a network of specialized agents available on Agentverse.

Overview

ASI:One's agentic models (asi1-agentic, asi1-fast-agentic, asi1-extended-agentic) are designed to automatically discover and coordinate with agents from the Agentverse marketplace to accomplish complex tasks. They handle agent selection, orchestration, and execution planning autonomously by connecting to the vast ecosystem of agents available on Agentverse.

Key Features:

Autonomous Agent Discovery: Automatically finds relevant agents from Agentverse marketplace for your tasks
Session Persistence: Maintains conversation context across multiple interactions
Asynchronous Processing: Handles long-running agent workflows from Agentverse
Streaming Support: Real-time response streaming for better UX

Quick Start

Python
cURL
JavaScript

import os, uuid, json, requests, sys, readline
import time

API_KEY = os.getenv("ASI_ONE_API_KEY") or "sk-REPLACE_ME"
ENDPOINT = "https://api.asi1.ai/v1/chat/completions"
MODEL = "asi1-fast-agentic"
TIMEOUT = 90  # seconds

# In-memory session management
SESSION_MAP: dict[str, str] = {}

def get_session_id(conv_id: str) -> str:
    """Return existing session UUID for this conversation or create a new one."""
    sid = SESSION_MAP.get(conv_id)
    if sid is None:
        sid = str(uuid.uuid4())
        SESSION_MAP[conv_id] = sid
    return sid

def ask(conv_id: str, messages: list[dict], *, stream: bool = False) -> str:
    """Send messages list to asi1-agentic; return assistant reply."""
    session_id = get_session_id(conv_id)
    print(f"[session] Using session-id: {session_id}")

    headers = {
        "Authorization": f"Bearer {API_KEY}",
        "x-session-id": session_id,
        "Content-Type": "application/json",
    }
    
    payload = {
        "model": MODEL,
        "messages": messages,
        "stream": stream,
    }

    if not stream:
        resp = requests.post(ENDPOINT, headers=headers, json=payload, timeout=TIMEOUT)
        resp.raise_for_status()
        return resp.json()["choices"][0]["message"]["content"]

    # Streaming implementation
    with requests.post(ENDPOINT, headers=headers, json=payload, timeout=TIMEOUT, stream=True) as resp:
        resp.raise_for_status()
        full_text = ""
        for line in resp.iter_lines(decode_unicode=True):
            if not line or not line.startswith("data: "):
                continue
            line = line[len("data: "):]
            if line == "[DONE]":
                break
            try:
                chunk = json.loads(line)
                choices = chunk.get("choices")
                if choices and "content" in choices[0].get("delta", {}):
                    token = choices[0]["delta"]["content"]
                    sys.stdout.write(token)
                    sys.stdout.flush()
                    full_text += token
            except json.JSONDecodeError:
                continue
        print()
        return full_text

# Simple usage example - agent will be called from Agentverse marketplace
conv_id = str(uuid.uuid4())
messages = [{"role": "user", "content": "Help me book a restaurant for dinner tonight"}]
reply = ask(conv_id, messages, stream=True)
print(f"Assistant: {reply}")

# Generate a session ID
SESSION_ID=$(uuidgen)

curl -X POST https://api.asi1.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $ASI_ONE_API_KEY" \
  -H "x-session-id: $SESSION_ID" \
  -d '{
    "model": "asi1-fast-agentic",
    "messages": [
      {"role": "user", "content": "Help me book a restaurant for dinner tonight"}
    ],
    "stream": false
  }'

import { v4 as uuidv4 } from 'uuid';

const API_KEY = process.env.ASI_ONE_API_KEY;
const ENDPOINT = "https://api.asi1.ai/v1/chat/completions";
const MODEL = "asi1-fast-agentic";

// Session management
const sessionMap = new Map();

function getSessionId(convId) {
    let sessionId = sessionMap.get(convId);
    if (!sessionId) {
        sessionId = uuidv4();
        sessionMap.set(convId, sessionId);
    }
    return sessionId;
}

async function ask(convId, messages, stream = false) {
    const sessionId = getSessionId(convId);
    console.log(`[session] Using session-id: ${sessionId}`);

    const response = await fetch(ENDPOINT, {
        method: 'POST',
        headers: {
            'Authorization': `Bearer ${API_KEY}`,
            'x-session-id': sessionId,
            'Content-Type': 'application/json'
        },
        body: JSON.stringify({
            model: MODEL,
            messages: messages,
            stream: stream
        })
    });

    if (!stream) {
        const result = await response.json();
        return result.choices[0].message.content;
    }

    // Handle streaming response
    const reader = response.body.getReader();
    const decoder = new TextDecoder();
    let fullText = '';

    while (true) {
        const { done, value } = await reader.read();
        if (done) break;

        const chunk = decoder.decode(value);
        const lines = chunk.split('\n').filter(line => line.trim());

        for (const line of lines) {
            if (line.startsWith('data: ')) {
                const data = line.slice(6);
                if (data === '[DONE]') return fullText;

                try {
                    const parsed = JSON.parse(data);
                    const content = parsed.choices?.[0]?.delta?.content;
                    if (content) {
                        process.stdout.write(content);
                        fullText += content;
                    }
                } catch (e) {
                    // Skip malformed JSON
                }
            }
        }
    }
    console.log();
    return fullText;
}

// Usage example - agent will be called from Agentverse marketplace
const convId = uuidv4();
const messages = [{"role": "user", "content": "Help me book a restaurant for dinner tonight"}];
const reply = await ask(convId, messages, true);
console.log(`Assistant: ${reply}`);

Session Management

Agentic models require session persistence to maintain context across agent interactions with the Agentverse marketplace. Always include the x-session-id header:

import uuid

# Create or retrieve session ID for conversation
def get_session_id(conversation_id: str) -> str:
    # In production, store this in Redis or database
    session_id = SESSION_MAP.get(conversation_id)
    if not session_id:
        session_id = str(uuid.uuid4())
        SESSION_MAP[conversation_id] = session_id
    return session_id

# Include in every request
headers = {
    "Authorization": f"Bearer {API_KEY}",
    "x-session-id": get_session_id("user_123_chat"),
    "Content-Type": "application/json"
}

Asynchronous Agent Processing

When agents from Agentverse marketplace need time to complete tasks, the model may send a deferred response. Poll for updates:

Python
JavaScript

def poll_for_async_reply(conv_id: str, history: list[dict], *, wait_sec: int = 10, max_attempts: int = 30):
    """Poll the conversation every `wait_sec` seconds for a deferred reply."""
    for attempt in range(max_attempts):
        time.sleep(wait_sec)
        print(f"🔄 polling (attempt {attempt + 1}) …", flush=True)
        
        reply = ask(conv_id, history, stream=False)
        if reply and "no new message" not in reply.lower():
            return reply
    return None

# Usage after getting "I've sent the message" response from Agentverse agent
if "I've sent the message" in assistant_reply:
    follow_up = poll_for_async_reply(conv_id, conversation_history)
    if follow_up:
        print(f"Agentverse agent completed task: {follow_up}")
        conversation_history.append({"role": "assistant", "content": follow_up})

async function pollForAsyncReply(convId, history, waitSec = 10, maxAttempts = 30) {
    for (let attempt = 0; attempt < maxAttempts; attempt++) {
        await new Promise(resolve => setTimeout(resolve, waitSec * 1000));
        console.log(`🔄 polling (attempt ${attempt + 1}) …`);
        
        const reply = await ask(convId, history, false);
        if (reply && !reply.toLowerCase().includes("no new message")) {
            return reply;
        }
    }
    return null;
}

// Usage
if (assistantReply.includes("I've sent the message")) {
    const followUp = await pollForAsyncReply(convId, conversationHistory);
    if (followUp) {
        console.log(`Agentverse agent completed task: ${followUp}`);
        conversationHistory.push({"role": "assistant", "content": followUp});
    }
}

Available Agentic Models

Model	Best For	Latency	Context Window
`asi1-agentic`	General orchestration & prototyping	Medium	32K tokens
`asi1-fast-agentic`	Real-time agent coordination	Ultra-fast	24K tokens
`asi1-extended-agentic`	Complex multi-stage workflows	Slower	64K tokens

Complete Interactive Example

Python CLI

import os, uuid, json, requests, sys, time

API_KEY = os.getenv("ASI_ONE_API_KEY") or "sk-REPLACE_ME"
ENDPOINT = "https://api.asi1.ai/v1/chat/completions"
MODEL = "asi1-fast-agentic"
TIMEOUT = 90

SESSION_MAP: dict[str, str] = {}

def get_session_id(conv_id: str) -> str:
    sid = SESSION_MAP.get(conv_id)
    if sid is None:
        sid = str(uuid.uuid4())
        SESSION_MAP[conv_id] = sid
    return sid

def _post(payload: dict, session_id: str, stream: bool = False):
    headers = {
        "Authorization": f"Bearer {API_KEY}",
        "x-session-id": session_id,
        "Content-Type": "application/json",
    }
    return requests.post(ENDPOINT, headers=headers, json=payload, timeout=TIMEOUT, stream=stream)

def ask(conv_id: str, messages: list[dict], *, stream: bool = False) -> str:
    session_id = get_session_id(conv_id)
    print(f"[session] Using session-id: {session_id}")

    payload = {
        "model": MODEL,
        "messages": messages,
        "stream": stream,
    }

    if not stream:
        resp = _post(payload, session_id)
        resp.raise_for_status()
        return resp.json()["choices"][0]["message"]["content"]

    # Streaming
    with _post(payload, session_id, stream=True) as resp:
        resp.raise_for_status()
        full_text = ""
        for line in resp.iter_lines(decode_unicode=True):
            if not line:
                continue
            if line.startswith("data: "):
                line = line[len("data: ") :]
            if line == "[DONE]":
                break
            try:
                chunk = json.loads(line)
            except json.JSONDecodeError:
                continue

            choices = chunk.get("choices")
            if not choices:
                continue

            delta = choices[0].get("delta", {})
            if "content" in delta:
                token = delta["content"]
                sys.stdout.write(token)
                sys.stdout.flush()
                full_text += token
        print()
        return full_text

def poll_for_async_reply(conv_id: str, history: list[dict], *, wait_sec: int = 10, max_attempts: int = 30):
    for attempt in range(max_attempts):
        time.sleep(wait_sec)
        print(f"\n🔄 polling (attempt {attempt + 1}) …", flush=True)
        reply = ask(conv_id, history, stream=False)
        if reply and "no new message" not in reply.lower():
            return reply
    return None

# Interactive CLI
if __name__ == "__main__":
    conv_id = str(uuid.uuid4())
    history: list[dict] = []

    print("Agentic LLM demo. Type Ctrl+C to exit.\n")
    try:
        while True:
            user_input = input("you > ").strip()
            if not user_input:
                continue
                
            history.append({"role": "user", "content": user_input})
            reply = ask(conv_id, history, stream=True)
            history.append({"role": "assistant", "content": reply})

            if "I've sent the message" in reply:
                follow = poll_for_async_reply(conv_id, history)
                if follow:
                    print(f"\n[Agentverse agent reply]\n{follow}")
                    history.append({"role": "assistant", "content": follow})
    except KeyboardInterrupt:
        print("\nBye!")

Best Practices

Session Management

Use UUIDs for session IDs to avoid collisions
Store session mappings in Redis or database for production
Include x-session-id header in every request to maintain context

Error Handling

Implement timeouts for long-running agent tasks
Handle network failures with exponential backoff
Validate responses before processing agent results

Performance Optimization

Use streaming for better user experience
Choose appropriate model based on complexity needs
Implement async polling for deferred agent responses

Agent Coordination

Be specific in requests to help agent discovery from Agentverse marketplace
Allow time for complex multi-agent workflows involving Agentverse agents
Monitor session state to understand Agentverse agent progress

Overview​

Quick Start​

Session Management​

Asynchronous Agent Processing​

Available Agentic Models​

Complete Interactive Example​

Best Practices​

Session Management​

Error Handling​

Performance Optimization​

Agent Coordination​

Overview

Quick Start

Session Management

Asynchronous Agent Processing

Available Agentic Models

Complete Interactive Example

Best Practices

Session Management

Error Handling

Performance Optimization

Agent Coordination