agloom 0.1.73

Production agent framework on LangChain/LangGraph: nine execution patterns, persistent memory, skills, feedback, multi-level HITL, MCP, AGP protocol, runtime bridge, and observability hooks.

agloom

Build agents that route themselves.
One familiar API — classification, memory, streaming, guardrails, and learning included.

Nine execution patterns. Auto-selected per task. Skills improve over time.

Documentation · Quick Start · PyPI · Examples · Issues

Start here

agloom is a Python framework for production-minded agents on LangChain / LangGraph. You describe the model and tools; agloom picks how to run the task (single-shot, ReAct, supervisor-style delegation, pipelines, and more), tracks steps and tokens, and can learn reusable skills from what worked.

If you already use LangChain’s agent APIs, think of create_agent as your main entrypoint — with orchestration, memory, streaming, and safety knobs in one place.

Install

pip install agloom
# optional extras, e.g. Groq:
pip install agloom[groq]

Your first agent

import asyncio
from langchain_groq import ChatGroq
from agloom import create_agent

async def main():
    llm = ChatGroq(model="meta-llama/llama-4-scout-17b-16e-instruct")
    agent = await create_agent(model=llm, name="my-agent")
    result = await agent.ainvoke("What causes auroras?")
    print(result.output)

asyncio.run(main())

create_agent is async (use await). From synchronous code, use create_agent_sync.

Next steps: Why agloom? · Patterns explained · All parameters

What you get (in plain language)

You want to…	agloom helps by…
Ship faster	Picking a strategy per query instead of hand-writing routers and graphs
Keep context	Session memory by default; optional long-term memory and skills
Show progress	Token streaming plus structured events for “thinking” / tool UIs
Stay safe	Human-in-the-loop levels, timeouts, retries, rate limits — configurable
Improve over time	Skill library and feedback hooks so behavior compounds

For the full feature tour, see What you get in the docs — the README stays short on purpose.

agloom CLI & web workspace

• Terminal: the agloom CLI (npm agloom-cli, repo agloom_cli/) is the terminal client — React-based UI. From that folder: npm install → npm run build → npm start. It talks to agloom-runtime over AGP (stdio by default). CLI quick start
• Browser: agloom_web/ is the Vite workspace for sessions and observability — same idea, run commands inside that folder.

PyPI’s agloom package includes the library and agloom-runtime. The agloom command prints a short pointer to the agloom CLI (repo folder agloom_cli/) for backwards compatibility.

Learn more (documentation hub)

Guide	What it’s for
Quick Start	Smallest path to a running agent
Execution patterns	How routing works (conceptual + diagrams)
Streaming & events	Responsive UI patterns
Production	Deploying, testing, operating
Errors & fixes	When something goes wrong

Requirements

• Python 3.12.x (see pyproject.toml on GitHub for the exact pin)
• Node.js ≥ 24.15.0 — only if you hack on agloom_cli/ or agloom_web/
• An LLM API key (Groq, OpenAI, NVIDIA, Hugging Face, or another LangChain-compatible provider)

Contributing & license

Contributions welcome — see CONTRIBUTING.md.

Licensed under Apache 2.0.

agloom is built by MEDHIRA

[e-mail] · GitHub · PyPI

_{Founded by S Muni Harish}