The Autonomous Company
Operating System
Maximum entropy. Full context. Let the agents run free.
The best results come from giving a language model everything — full context, every secret, every integration, every piece of institutional knowledge — and letting it operate freely. For that, it needs a proper machine. A real computer running 24/7 where all the context is stored, all the credentials live, all the work accumulates, and all the other agents work alongside it.
A Kortix is a company. One shared machine where every agent sees the same filesystem, the same databases, the same history. Context isn't siloed per tool or per session — it's shared across the entire system, compounding over time. Coding agents operating inside a full Linux environment are the optimal harness for all knowledge work — not just engineering, but sales, finance, ops, legal, support, and everything in between. We're building this as our own internal operating system to run our own companies, and open-sourcing the whole thing.
The system
Everything runs inside one shared Linux Ubuntu machine. Agents have bash, a real filesystem, package managers, databases, and the entire software ecosystem — plus full access to every credential, file, and piece of context in the system. One machine, total openness, maximum results.
Markdown files with their own identity, permissions, tools, and triggers. A support agent, a bookkeeper, a recruiter, a sales rep — each a specialist operating inside the same machine.
Reusable knowledge packs that teach agents how to do real work — coding, browser automation, deep research, legal writing, spreadsheets, data analysis, and more. 60+ built-in, and writing new ones is just writing code.
The autonomous execution loop. An agent works until the task is done, self-verifies, and only stops when it can prove the result is correct.
Time-driven or event-driven. Cron schedules and webhooks with prompt, command, or HTTP actions. Git-versionable config. Morning briefings, recurring jobs, real-time reactions.
Persistent, filesystem-based, and shared across all agents. Every decision, preference, and context is retained on the same machine. The longer the system runs, the smarter the whole company gets.
One agent delegates to many. Projects, background sessions, parallel sub-agents. A primary orchestrator decomposes work and tracks it to completion — like departments in a company.
Runs on OpenCode — an open foundation for agent skills, tools, and commands.
How it works
Connect your tools. Configure your agents. Deploy them. An entire company's knowledge work — automated inside one machine.
Every tool your company uses — OAuth apps, MCP servers, REST APIs, CLI tools, environment variables. If it has an interface, Kortix connects to it. 3,000+ integrations available, and custom ones are trivial to add.
3,000+ via OAuth · MCP · REST · CLI · env vars
Define agents for every function — engineering, sales, finance, ops, legal, support. Each agent is a markdown file with its own identity, permissions, skills, and activation rules. Because they run inside a full Linux environment, anything you can script, they can do. Compose them into an autonomous workforce.
Markdown files that define autonomous workers — identity, permissions, tools, triggers. Each one a specialist.
Knowledge packs that teach agents what to do — coding, research, browser automation, legal writing, spreadsheets. 60+ built-in.
Slash commands that trigger structured workflows. /autowork and /onboarding — your playbooks, automated.
Cron schedules and webhooks defined in agent frontmatter. Morning briefings, event-driven reactions, recurring jobs.
Your agents run 24/7. Triggers fire them on schedule. Autowork keeps them going until the job is verified done. You check in from the dashboard, your phone, or your team's messaging platform — whenever you want.