you can now connect any mcp server
Add a custom MCP server to opencompany from settings. Its tools become actions your agents can call in chat, workflows, and coding tasks, with per-tool permissions and encrypted personal credentials.
Product
Blog
Launching the public beta of opencompany
opencompany is a model-independent, open-source AI workspace for small teams. Use any model, keep your agents running in the cloud, ship code with Claude Code and Codex in isolated sandboxes, and build shared context in a built-in wiki.
Company
The execution substrate is the product
What Grok Bot's architecture teaches you about building persistent agents.
Engineering
Light Work and Deep Work: How We Route AI Agent Tasks
Giving an AI assistant a lot of tools without making it slow, expensive, or dumb is the core problem of agent design. Here's why we split it into light work and deep work — and route each to the engine that's actually good at it.
Engineering
Running AI Agents in Production: What Changes When It's Not a Demo
Most AI agent tutorials skip the hard parts. Here's what actually matters when you move agents from a laptop to a team — permissions, secrets, audit trails, and the infrastructure decisions that shape everything.
Engineering
OpenClaw on Mac Mini: The Security-First Setup Guide
42,665 OpenClaw instances found exposed online. Here's how to set up OpenClaw on a Mac Mini with security as priority one -- from an agent security team.
Guides
OpenClaw: Mac Mini vs Cloud — Which Deployment Is Right for Your Team?
Mac Mini at $599 or a $20/mo VPS? We break down the real costs, security tradeoffs, and scaling limits of every OpenClaw deployment option.
Guides
Exploring OpenClaw Alternative for Teams That Need Security Controls
OpenClaw is powerful but has 9 CVEs, supply chain attacks, and no permission model. Here are the alternatives that solve security for team deployments.
Security
OpenClaw Security: What You Need to Know Before Deploying to Your Team
OpenClaw has 9 CVEs and a supply chain attack. Snyk found security flaws in 36.82% of scanned agent skills. Here's what teams need to know.
Security
Permission Models for AI Agents: Off, On, and Ask
Giving an AI agent full access is easy. Giving it the right access is the actual problem. Here's how permission models work, why most tools get them wrong, and what a production-grade approach looks like.
Security