kortix-ai/CompanyOS
AI EngineeringAn open-source platform for running custom AI agents that handle scheduled, event-driven, multi-agent company work on isolated cloud computers.
What it does
Kortix CompanyOS is a cloud-native agent harness for a whole company: you define agents, skills, rules, and connectors as files in one git repo, and a dispatcher called My Kortix routes each task to a purpose-built agent session. Each session runs on its own isolated cloud computer, sleeps when idle, and wakes with its full history.
The work can be started by schedules, signed webhooks, connector or app events, and merged change requests. Sessions can hand off work in the background, run workflows that coordinate many agents, message each other, and report results back. Every change that lands on main goes through a change request a person approves.
It is aimed at teams that want shared, versioned agents and company context rather than one-off chat sessions. Agents can use 3,000+ app connectors plus MCP, OpenAPI, GraphQL and raw HTTP, and can build and host internal apps and sites behind Kortix sign-in. Models are pluggable through the Kortix LLM gateway, with managed models, your own provider keys, or ChatGPT/Claude subscriptions.
Quickstart
Prepare a shell environment and install the CLI:
curl -fsSL https://kortix.com/install | bash
Then scaffold a project and ship it:
kortix init
kortix ship
kortix init generates kortix.yaml plus the agents, skills and runtime configuration; kortix ship pushes the repo and brings the setup live in the cloud. Afterwards you can create a session from a prompt, review changes and chat from the terminal:
kortix sessions new --prompt "Summarize this week's commits and open a change request"
kortix cr ls
kortix chat
If you prefer not to set up anything locally, you can create an account, a project and a session in the hosted app; the Guide covers the full workflow. The commands and their order follow the Quickstart.
Self-hosting
A self-hosted installation is one Docker Compose stack in your cloud account, VPC or on-prem servers, using the same images as the managed cloud:
curl -fsSL https://kortix.com/install | bash
kortix self-host start
kortix hosts use selfhost
The first run asks six questions and no model key; you connect GitHub and add model keys in the app afterwards. It is a self-hosted install, not a disconnected one: agent sandboxes run on the configured compute provider (Platinum by default) and the stack pulls images from Docker Hub. See the self-hosting guide for details.
Constraints and licensing
- Platform and prerequisites: the quickstart targets a Unix shell with
curlandbash; local development usespnpmand the repository's Docker environment. Apps are split intoapps/web,api,cli,desktop-electron,mobileandsandbox. - External dependencies: self-hosting still depends on a configured sandbox compute provider and image registry, so it is not an air-gapped deployment.
- Access and data: connector credentials are held server-side, and secrets are encrypted at rest per project. Roles, groups, per-resource permissions, audit logs, SAML 2.0 SSO and SCIM 2.0 are part of the team controls. SOC 2 Type I is reported for the cloud offering; verify what applies to your deployment.
- License: the repository metadata does not identify a standard OSI license (NOASSERTION). A separate white-label license lets you rebrand and host it, on the condition that Platinum is used as the exclusive compute provider for everything it offers (WHITE-LABEL-LICENSE.md). Read the actual license files before commercial use.
HotRepo's take
In the 2026-11-08 TypeScript daily ranking, the project recorded 20,552 stars (up 92 that day) and 3,459 forks, with a best position of 5 on that language list. That places it among the more visible TypeScript projects on that day, but the single snapshot cannot establish a trend, and no all-language daily rank is recorded. The architecture is the interesting part: separating agent state from disposable cloud computers, triggering sessions from events, and requiring human-approved change requests for anything reaching main are concrete choices that distinguish it from a chat-session agent wrapper. Whether the self-hosted path delivers on the sovereignty claims is the thing to check against your own infrastructure and provider setup.
Written by AI from the project's README, updated . The project's own docs are the source of truth.