OpenCompany

OpenCompany is a self-hosted canvas for AI agent workflows. You drag nodes onto a canvas, connect them, and give an agent memory and skills; pressing Run tests a node in place, while Start deploys the whole workflow as a durable background listener that waits for email, messages, or a schedule. It ships 146 nodes across 31 categories, and three example workflows load on first launch. It runs on Node.js 22 and Python 3.12 with the model keys you supply, or against local models through Ollama or LM Studio.

Twenty agent node types cover different jobs: general-purpose AI and Chat agents, AI Employee and Orchestrator team leads that assign bounded work through a built-in task manager, and specialists for Android phone control, browser automation, coding, Google Workspace, social messaging, scheduling, travel, payments, and customer support. Agents wrapping Claude Code and OpenAI Codex are included. Skills are short markdown files that teach an agent when to use which tool, editable in the interface, with 77 built in across 19 folders and user files in .opencompany/skills/ overriding built-ins of the same name.

Connecting a Context node stores an agent's conversation durably per workflow generation, so every trigger firing continues the same thread; at 80% of the model's context limit the agent asks the selected model for a five-section summary and continues from it. Execution runs on Temporal, with bounded retries for ordinary activities, once-only billed LLM steps, and a local executor when it is disabled. Credentials live in a separate database encrypted with Fernet and a PBKDF2-SHA256 key at 600,000 iterations, and single-owner JWT or multi-user auth can be turned on for shared deployments.

Provider	Notes
OpenAI	GPT-5.6 Sol / Terra / Luna (+ Pro variants), GPT-5.5, GPT-4.1
Anthropic	Claude Opus 5, Fable 5, Sonnet 5, Opus 4.8 / 4.7, with extended thinking
Google	Gemini 3.6 / 3.5 Flash, 3.1 Pro, with reasoning budgets
xAI	Grok 4.20, 4.20 multi-agent, 4.3, selected from the agent's provider dropdown
DeepSeek	DeepSeek V4 Flash / Pro
Kimi	Kimi K3
Mistral	Mistral Large / Medium / Small, Codestral
Groq, Cerebras	GPT-OSS-120b and others
Sarvam	Indic-first models (sarvam-105b, 128K context)
OpenRouter	200+ models through one API
Ollama, LM Studio	any local model, context length detected from the running server

Features

• Node canvas: 146 nodes across 31 categories, drag-to-map outputs onto another node's inputs, live execution animations, and a chat, console, and terminal panel
• Durable agent teams: an AI Employee or Orchestrator connects specialists through input-teammates and queues bounded tasks, running up to three descendants in parallel with lead acceptance and a read-only team monitor
• Messaging and mail: Gmail, Microsoft 365 over Graph, IMAP polling triggers, WhatsApp personal and Business Cloud API, Telegram, Discord, and Twitter/X, normalized by a unified social node
• Phone control: pair an Android device by QR code for battery, network, app launching, radios, camera, sensors, and media playback across 16 device services
• Web automation: an accessibility-tree browser, an alpha harness driving real Chrome over CDP, Crawlee and Apify scraping, and search through DuckDuckGo, Brave, Serper, or Perplexity
• Code and deployment: sandboxed Python, JavaScript, and TypeScript workspaces, a process manager for dev servers, and GitHub, Vercel, Cloudflare, and Google Cloud nodes authenticated through their own CLIs
• Retrieval: PDF and HTML parsing, chunking, local or OpenAI embeddings, and storage in ChromaDB, Qdrant, or Pinecone, queryable from any agent
• Cost tracking: USD cost calculated from provider-reported usage on memory-connected runs, with custom pricing in pricing.json