noesium 0.3.4

Towards a cognitive agentic framework

Noesium is a computation-driven cognitive agentic framework. Use it as a ready-to-run multi-agent system (NoeAgent plus subagents) or as a foundation to build custom agents with core abstractions and 17+ toolkits.

Design Philosophy

Noesium follows an event-sourced multi-agent kernel architecture for durability, replayability, and distributed coordination:

• Single execution authority — One Agent Kernel per agent; all reasoning, planning, tool use, and delegation happen inside it. No external layer mutates agent state.
• Event-sourced state — State is derived from an append-only event log. Enables replay, deterministic reconstruction, audit, and crash recovery.
• Delegation via events — Agents publish task events to capability topics; subscribers process them. Loose coupling, capability-based coordination.
• Separation of concerns — Cognition (Kernel), transport (Event Bus), persistence (Event Store), memory (Projection), routing (Topic subscriptions).

Goals: long-running autonomous agents, durable/resumable execution, multi-agent collaboration, horizontal scalability.

Architecture

flowchart TB
  subgraph framework [Framework]
    core[core: agent base, llm, toolify, memory, event, vector_store]
    toolkits[toolkits: 17+ built-in]
  end
  subgraph builtin [Built-in multi-agent system]
    noeAgent[NoeAgent]
    subagents[BrowserUseAgent, TacitusAgent, AskuraAgent]
  end
  framework --> builtin
  framework -.-> yourAgents[Your custom agents]

• Framework — Core modules (agent base classes, llm, toolify, memory, event, vector_store) and 17+ toolkits provide the agentic building blocks.
• Built-in multi-agent system — NoeAgent (orchestrator) and subagents (BrowserUseAgent, TacitusAgent, AskuraAgent) are a reference implementation and directly usable. See AGENTS.md for details.

Install

pip install -U noesium

Recommended for full features:

pip install noesium[all]

Optional extras:

pip install noesium[llm]             # OpenAI, LiteLLM, Instructor
pip install noesium[local-llm]       # Ollama, LlamaCPP
pip install noesium[agents]          # LangChain, LangGraph
pip install noesium[tools]           # 17+ toolkits
pip install noesium[browser-use]     # Browser automation

Use uv run for scripts when developing with the repo.

Quick Start

1. Common user — NoeAgent

Library:

import asyncio
from noesium import NoeAgent

async def main():
    agent = NoeAgent()
    result = await agent.arun("What are the latest developments in quantum computing?")
    print(result)

asyncio.run(main())

TUI:

python -m noesium.noeagent.tui

Full NoeAgent options, streaming, subagents, and config: docs/noeagent_guide.md.

2. Developer — Framework

Build a custom agent on the framework:

from noesium.core.agent import BaseGraphicAgent
from noesium.core.llm import get_llm_client
from noesium.core.toolify import get_toolkit

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

    def build_graph(self):
        # Define your LangGraph workflow
        pass

agent = MyAgent()
result = await agent.run("Your task")

For full framework usage, custom agents, and toolkits: docs/dev_guide.md.

Environment variables

export NOE_LLM_PROVIDER="openai"   # or openrouter, ollama, llamacpp
export OPENAI_API_KEY="sk-..."
export SERPER_API_KEY="..."        # for search toolkit

License and support

• License — MIT.
• Documentation — AGENTS.md, docs/dev_guide.md, docs/specs/, examples/.
• Issues and repo — GitHub.