Noesium is a computation-driven cognitive agentic framework providing foundational abstractions for building autonomous AI agents with planning, memory, tools, and orchestration capabilities.

Design Philosophy

Noesium is built on an Event-Sourced Multi-Agent Kernel Architecture that prioritizes durability, replayability, and distributed coordination. The core principles include:

Single Execution Authority

Each agent contains exactly one execution authority: the Agent Kernel. All reasoning, planning, branching, retries, tool invocation, and delegation logic occur inside this kernel. No external orchestration layer is permitted to mutate agent state.

Event-Sourced State

State is not stored as mutable objects. Instead, every state transition emits an event, the event log is the source of truth, and current state is a projection derived from event history. This ensures replayability, deterministic reconstruction, auditable cognition, and crash recovery.

Delegation via Events

Delegation is modeled as event emission, not direct invocation. An agent expresses intent by publishing a task event to a capability topic. Any agent subscribed to that topic may process the task, enforcing loose coupling and capability-based coordination.

Separation of Concerns

• Cognition → Agent Kernel
• Transport → Event Bus
• Persistence → Event Store
• Memory Projection → Projection Layer
• Capability Routing → Topic Subscriptions

Architectural Goals

The system supports long-running autonomous agents, infinite reasoning loops, durable/resumable execution, distributed multi-agent collaboration, deterministic replayability, horizontal scalability, and explicit separation of cognition, transport, and persistence.

Installation

pip install -U noesium

Optional Dependencies

# Recommended: all features
pip install noesium[all]

# AI providers
pip install noesium[llm]             # OpenAI, LiteLLM, Instructor
pip install noesium[local-llm]       # Ollama, LlamaCPP
pip install noesium[ai-providers-all] # All AI providers

# Frameworks
pip install noesium[agents]          # LangChain + Bubus

# Tools & data
pip install noesium[tools]           # 17+ toolkits
pip install noesium[datascience]     # Pandas, NetworkX
pip install noesium[browser-use]     # Browser automation

Quick Start

LLM Client

from noesium.core.llm import get_llm_client

# Create client (supports openai, openrouter, ollama, llamacpp)
client = get_llm_client(provider="openai", api_key="sk-...")

# Chat completion
response = client.completion([{"role": "user", "content": "Hello!"}])

# Structured output
from pydantic import BaseModel

class Answer(BaseModel):
    text: str
    confidence: float

client = get_llm_client(provider="openai", structured_output=True)
result = client.structured_completion(messages, Answer)

Build an Agent

from noesium.core.agent.base import BaseGraphicAgent
from noesium.core.llm import get_llm_client

class MyAgent(BaseGraphicAgent):
    def __init__(self, llm_client=None):
        super().__init__(llm_client or get_llm_client())

    def build_graph(self):
        # Define your agent's workflow graph
        pass

agent = MyAgent()
result = await agent.run("What is the meaning of life?")

Use Toolkits

from noesium.core.toolify import get_toolkit, ToolkitConfig

# Search toolkit
search_config = ToolkitConfig(name="search", config={"SERPER_API_KEY": "..."})
search = get_toolkit("search", search_config)
results = await search.search_google_api("Python async programming")

# Bash toolkit
bash = get_toolkit("bash")
files = await bash.list_directory(".")

Core Modules

Module	Description
Agents (noesium.core.agent)	BaseAgent, BaseGraphicAgent, BaseHitlAgent, BaseResearcher
LLM (noesium.core.llm)	Multi-provider support, structured output, token tracking
Toolify (noesium.core.toolify)	Unified tool system with 17+ built-in toolkits
Memory (noesium.core.memory)	Multi-tier memory with semantic search
Event (noesium.core.event)	Domain events, message bus, tracing
Vector Store (noesium.core.vector_store)	PGVector, Weaviate support

Built-in Agents

• Noe - Autonomous research assistant with planning, execution, reflection cycles
• BrowserUseAgent - Web automation with DOM interaction and code execution
• TacitusAgent - Advanced research agent with iterative query generation, web search, reflection, and answer synthesis

Built-in Toolkits

Search, Bash, Memory, Python Executor, ArXiv, Audio, Document, File Edit, GitHub, Gmail, Image, Tabular Data, Video, Wikipedia, and more.

See AGENTS.md for detailed documentation.

Environment Variables

# LLM Provider
export NOESIUM_LLM_PROVIDER="openai"
export OPENAI_API_KEY="sk-..."

# Search
export SERPER_API_KEY="..."

Documentation

• AGENTS.md - Detailed agent and toolkit documentation
• specs/ - Design specifications (RFCs)
• examples/ - Usage examples

License

MIT License