Own the stack
Self-hosting a Claude Cowork alternative.
Self-hosting an agent platform is not the same as installing a desktop app. It means the agent runtime, your data, and your configuration live on hardware you control — and that a human still reviews what the agents produce.
The basics
What "self-hosted" actually means here
- The runtime lives on your machine. Agents run in an environment you provision — a laptop, a VPS, your own VPC, or an on-prem network — not on a vendor's cloud.
- Your configuration is code you own. Agents, skills, memory, and connectors are files in a repository you version and diff, rather than settings in someone else's product.
- Secrets and data don't leave your perimeter. Credentials are stored where you control them, and model calls go to providers you've chosen with your own keys.
Why this matters: Claude Cowork runs in Anthropic's cloud on Anthropic's models. A self-hosted alternative moves all three layers — runtime, configuration, and data — back under your control, at the cost of you operating the infrastructure.
Kortix (Suna)
The Docker path
Kortix is the most server-oriented of the three: it runs thousands of isolated sessions in parallel and lands work through change requests — a closer fit for an org-scale fleet than a single desktop.
The first interactive setup asks only for the integration credentials that unlock managed git, GitHub access, and connectors — ports, local URLs, keys, and Docker Compose defaults are generated for you. Note that self-host start pulls its images from Docker Hub, so this is a self-hosted install rather than a fully air-gapped one.
Every session runs in an isolated sandbox on its own branch. The agent commits and pushes; you review the change request and merge to main. Nothing reaches production unreviewed.
License caveat (honest): Kortix is Elastic License 2.0 — source-available, not OSI-open. You can read, fork, and self-host it, but you can't offer it as a competing hosted service. If you need OSI-open, weigh OpenWork or Eigent instead.
Alternatives
Desktop-local and workspace options
OpenWork
Local-first by design: you install the desktop app, your files stay on your machine, and prompts go only to the LLM provider you choose. No server to operate — simpler, but no always-on agent fleet.
Eigent
Packages a workspace you run locally or self-host, connecting the models you choose (cloud, enterprise gateway, or local). Its review and approval flow keeps permission decisions visible.
Kortix (Suna)
Org-scale specialist agents running in parallel with a shared memory, work landing only through reviewed change requests, per-resource permissions, an audit trail, and encrypted-at-rest secrets.
When desktop-local beats server-side
- You're a solo operator or small team with no interest in running infrastructure.
- Your work is file-centric on a single machine (documents, spreadsheets, research).
- You want zero cloud footprint and are comfortable with one machine's uptime.
When server-side (Kortix) wins
- You need org-scale specialist agents running in parallel with a shared memory.
- You want work to land only through reviewed change requests.
- You need per-resource permissions, an audit trail, and encrypted-at-rest secrets.
Checklist
A self-hosting checklist
- Confirm the license fits your use (OSI-open vs source-available).
- Choose hardware: laptop, VPS, VPC, or on-prem.
- Bring your own model keys or point at local models — no Anthropic subscription needed.
- Scope secrets to the minimum; keep credentials out of the agent's reach except at runtime.
- Turn on the human gate: change requests or approvals, deny-by-default.
- Test one small workflow end-to-end before onboarding a team.