Catalog PyPI
PACKAGE EXPLORERView in registry

agentos-runtime

PyPI

AgentOS Runtime — AI Agent CLI for workplace automation

agentaiclienginellmmulti-agent-system(mas)osoutlookruntimeteams
Versions collected1
Direct dependencies7From release 0.1.0
Last observed20 Oct, 22:50 UTC

README

Version 0.1.0 documentation

AgentOS Runtime

AI Agent CLI for workplace automation — schedule meetings, generate workflows, and automate repetitive tasks using natural language.

pip install agentos-runtime
agentos init
agentos schedule "invite john@company.com to onboarding sync tomorrow at 2pm"
agentos workflow "Create a Workflow Definition for AgentOS Website to execute Enterprise Process Automation"

What it does

AgentOS Runtime gives you a terminal command (agentos) that uses an AI agent (ReAct loop with LLM function calling) to:

  • Schedule Teams meetings from plain English — the agent reasons about your request, generates a rich meeting agenda, opens Outlook, and sends the invite
  • Attach files (PDF, Word, invoices) to meeting invites
  • Generate agentic workflows for multi-step enterprise processes (requires AgentOS backend)

Requirements

  • Python 3.10+
  • Windows (Outlook COM automation requires Outlook desktop installed and logged in)
  • One of: Gemini API key, OpenAI API key, Anthropic API key, GCP service account, or any LiteLLM-supported provider

Install

pip install agentos-runtime

Setup

agentos init

The setup wizard asks you:

  1. Which LLM provider (OpenAI, Anthropic, Gemini, Vertex AI, Groq, Azure, Kimi...)
  2. Your API key or GCP service account path
  3. Your Outlook email (must be logged into Outlook desktop)
  4. AgentOS server URL (leave default if running locally)

Config is saved to ~/.agentos/config.json — your keys stay on your machine.


Usage

Schedule a Teams meeting

agentos schedule "invite sarah@company.com to 1:1 sync tomorrow at 3pm"

agentos schedule "onboarding kickoff for John on April 20 at 10am for 2 hours"

agentos schedule "PROCESS.AI discussion with team@co.com on 15/04 from 14:00 to 16:00" \
  -a project_brief.pdf \
  -a onboarding_checklist.docx

The agent will:

  1. Resolve the date and time
  2. Ask for any missing attendee emails via a popup
  3. Generate a full meeting agenda using the LLM
  4. Open Outlook and send the invite with all attachments

Generate a workflow

# Requires AgentOS backend running: python run.py
agentos workflow "onboard a new employee with HR, IT and procurement steps"
agentos workflow "laptop procurement with manager approval" --name laptop_flow

Manage config

agentos config show                          
agentos config set organizer_email x@co.com  
agentos config set server_url https://api.yourdomain.com

commands sheet

agentos --help

Supported LLM Providers

ProviderHow to configure
OpenAI (GPT-4o, GPT-4-turbo)API key from platform.openai.com
Anthropic (Claude Sonnet/Opus)API key from console.anthropic.com
Google GeminiAPI key from aistudio.google.com
Google Vertex AIGCP service account JSON + project ID
Groq (LLaMA, Mixtral)API key from console.groq.com
Kimi / MoonshotAPI key from platform.moonshot.cn
Azure OpenAIAPI key + endpoint URL

Switch provider anytime: agentos config set llm_provider openai


How it works

agentos schedule "..."
       │
       ▼
CLI loads ~/.agentos/config.json → sets env vars
       │
       ▼
ReAct Loop (LLM reasons + calls tools):
  1. get_current_date      → resolves relative dates
  2. ask_user_for_input    → popup for missing emails
  3. generate_agenda       → LLM writes meeting email body
  4. schedule_meeting      → Outlook COM creates + sends invite
       │
       ▼
Outlook opens live → invite delivered to attendees

License

MIT — see LICENSE

Downloads & popularity

Awaiting observations

No metric observations yet

Download counts and popularity will appear here as the collector receives them.

Versions 1

Collected releases
VersionPublishedDirect dependenciesOpen version
0.1.012 Mar 20267

Package advisories

0
No matching advisories collected

This describes our collected data; it is not a security assessment.

Repository

PENDING
mohithtummala/agentos-workflow-engine

Shared repository data · GitHub

License detected in repository
Not collected

Metadata history

Meaningful changes
Current collected state Current

Observed since 20 Oct, 22:50 UTC

No superseded metadata states collected. Unchanged observations do not create new revisions.

Record identity

package:b3c73dd6b515d1aebfffb2a9df1a849ece44f77a81ccd104ff50d10d50ceeb4e