Beyond Code Mode: The Agentica SDK
Build agents that interact with your runtime objects by writing code. State of the art performance, securely sandboxed.
Get Started Now# 3rd party API for fetching market datafrom third_party_library import Client1. Get any runtime object1. Get any runtime objectfrom agentica import spawn # Pass the client to the agentdata_agent = await spawn( scope= {'client': Client()}) # Ask it a question and print the resultprint(await data_agent.call( float, "How much have I spent this month?"))Build Code Mode Agents with Any Model
A Python and TypeScript SDK for building code mode agents that have access to your runtime objects. Deploy on the Symbolica Platform.
Get StartedOur open-source ARC-AGI-3 implementation solves all 3 publicly available pre-release games and our ARC-AGI-2 implementation achieves a score of 85.28% with Opus 4.6 (120k) High, increasing the scores of GPT 5.2 (XHigh) and Opus 4.5 by 10 and 20 percentage points respectively.
Runtime as Context: How Agentica Reasons Over Data
What changes when an agent has access to a persistent Python runtime instead of reconstructing context from prior tool outputs? To make this distinction clear, we built a data analysis demo comparing a traditional tool-calling agent built with LangChain’s Deep Agents, and Agentica, which reasons inside a Python runtime (REPL) where intermediate objects are retained across turns.
Watch a multi-agent system, built with Agentica
Query comes in: "Backtest a TSLA strategy." Agents jump into action fetching data, running calculations, generating charts seamlessly. Agentica’s types, composition, and dynamic code execution let tasks flow like one mind, orchestrating workflows in real time.
Code is the reasoning layer
Code is the reasoning layer
Objects in. Capabilities out.
A paradigm that goes beyond tool calls. Pass objects by reference in, specify a return type and get a runtime object back, types intact.
An agent with a persistent working memory that can think outside of the context window.
If you can import it, they can use it
Your code? SDK clients? APIs? No wrapping, just import and pass them in scope.
Agents that delegate capabilities
Agents can pass objects from their own scope to subagents — not just delegate tasks. This is autonomous orchestration.
Observability built for code execution
Every agent action is observable and traceable. See generated code, execution results, and spawn trees — rendered live, zero config.
from agentica import spawnfrom sdk import Database, UserIdagent = await spawn("Answer questions by searching the web.",model='openrouter:google/gemini-2.5-flash',scope={"database": Database(...)})await agent.call(dict[UserId, str],"For each user, summarise their spending habits.")
# 3rd party API for fetching market datafrom twelvedata import TDClientfrom agentica import spawn# Construct an instancemarket_data = TDClient(apikey=...)# Pass the client to the agentdata_agent = await spawn(premise="You are a financial data analyst.",model="anthropic/claude-opus-4.5",scope={'TDClient': TDClient, 'market_data': market_data})
[12:34:56] orchestrator → spawn researcher[12:34:57] researcher ← initialized[12:34:58] orchestrator → call research()[12:35:02] researcher → tool: web_search[12:35:05] researcher ← results: 15 items[12:35:06] orchestrator ← research complete[12:35:07] orchestrator → spawn writer[12:35:08] writer ← initialized[12:35:09] orchestrator → call write(research)[12:35:14] writer ← draft complete
Open-source framework, designed for the Symbolica platform
Symbolica
GPT-5.1
Claude Opus 4.5
Gemini 3 Pro
Opus 4.5
Agentica
Open-source framework
Agentica is open-source. Run locally or on the Symbolica platform.
Agentica SDKSymbolica platform
The Symbolica platform enables enterprise-scale agentic workloads today. Seamlessly integrate and manage multi-agent workloads with production applications.
Security, permissioning, telemetry and observability. Every agent action is observable and traceable in real-time.
Get started with Agentica.
In minutes.
Build multi-agent systems that can execute in your runtime today.
Get startedPowerful agents.
Simple pricing.
We charge a 5% fee on top of our underlying inference costs.
See OpenRouter for exact pricing.
5%
of model inference cost
Agentica, a new agent framework by Symbolica, is built on the premise that code is the most expressive interface through which models can interact with their environment.
Agents can interact with your runtime objects in a sandboxed execution environment. This includes the framework itself, allowing them to delegate capabilities, not just tasks, to other agents.