Best Claude Cowork alternatives in 2026: open source first

Claude Cowork is Anthropic's desktop agent for knowledge work: you point it at a folder and a goal, it plans the job, runs it, and hands back a finished result. It is good at that. Three walls still send people looking for alternatives, and the field's own guides name them plainly: the code is closed, the ecosystem is locked to one vendor, and only that vendor's models run (Eigent's 2026 guide).

An alternative earns a switch by removing at least one wall. The 2026 field splits on one more question that decides most of the rest: where the work runs. It can run on your machine, in a vendor's cloud, or on cloud computers you control, and that choice decides what the tool can do when you step away. Scheduled and background work is the clearest test: an agent that stops when the laptop sleeps is a desktop helper, and one that keeps working is a workforce (Kilo's 2026 guide).

Kortix: the recommendation

Kortix is the open-source AI Operating System: your agents, their skills, your company memory, and every connector in one git repo you own, with the agents working on real cloud computers. Open source: read it, fork it, audit it, with 20,000+ GitHub stars behind it as of October 2026 (Kortix on GitHub).

Model choice is as wide as this field gets: any model provider with your own keys. Or the ChatGPT plan you already pay for. Deployment covers all three options at once: run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free. Configuration lives in kortix.yaml and the files around it, versioned and reviewed like code, so you can grep your whole company.

Every session gets its own cloud computer: a disposable, isolated sandbox on its own branch. Session work reaches the default branch through a change request a person approves. Thousands of agents in parallel on one config, each on its own cloud computer. Reach is broad: 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP. Free is $0 with 200 credits per month and 1 project. The team plan is $40 per seat per month (Kortix).

OpenWork: the open-source desktop app

OpenWork is the free, open-source option that stays closest to Cowork's shape: a desktop app for macOS, Windows, and Linux, built on OpenCode, where agents do real work on your own files. It runs 50+ model providers on your own API keys, or local models through Ollama, and in desktop mode your files stay on your machine (OpenWork's site, its GitHub README). The desktop app costs nothing; OpenWork Cloud gives the first five seats free and then charges $10 per seat per month, and teams share skills and MCP servers from one place (checked October 2026). Two things it does not offer: Kortix's cloud, VPC, and on-prem range, and a git repo that holds the whole company's configuration, since the work lives on the machine you installed the app on.

Eigent: the multi-agent local platform

Eigent is an open-source, multi-agent workforce desktop built on the CAMEL-AI framework, which its own makers position as the open-source Cowork desktop with a deeper focus on parallel agents. It ships 200+ built-in MCP tools, runs a local backend on your machine so data stays local, takes your own API keys for Gemini and OpenAI-compatible providers, and adds SSO and role-based access for companies. Its own write-up names the trade-off: a backend with its own database and orchestrator takes more upkeep than a desktop app that just runs (Eigent's guide).

Claude Cowork and ChatGPT Work: the closed incumbents

Claude Cowork ships on Anthropic's paid plans, starting at $20 per month (checked October 2026). The desktop app can run sessions on your machine, and remote sessions run on Anthropic-managed infrastructure; either way, every step calls an Anthropic model, there is no self-host option, and the setup you build stays in their product (Nimbalyst's comparison). ChatGPT Work is the same shape on OpenAI's side: closed source, GPT-5.6 only, OpenAI's cloud, no self-host, with both incumbents tabulated side by side in Kortix on GitHub. Both are capable single-player tools for one task at a time; neither is built for a team that wants to own its configuration or run several agents at once.

The rest of the 2026 field

Four more names cover the edges of this field, each described from its own comparison post:

• Gumloop builds and runs AI workflows in the cloud, connects to any MCP server, and takes models from Anthropic, OpenAI, Meta, and open-weight providers; its Pro plan is $37 per month after a 14-day trial, with RBAC, SCIM/SAML, and VPC deployment on the enterprise tier (checked October 2026) (Gumloop's post).
• Vellum is an open-source personal AI assistant that carries one identity and persistent memory across Mac, iOS, web, email, Telegram, and Slack, with a separate credential process so the model requests actions without holding raw secrets, and free cloud hosting at signup (Vellum's review).
• Nimbalyst runs Claude Code and Codex side by side on your own provider accounts, puts sessions on a kanban board with optional worktree isolation, and keeps files in ordinary git folders on your machine; its individual-use features are open source, and an iOS app steers sessions from your phone (Nimbalyst's post).
• Kuse is a Rust-native, open-source cowork desktop that works directly on your local file system, with BYOK across OpenAI, Anthropic, and local models, and skills through MCP (Kuse's post).

Worth knowing as you read around: Gumloop, Vellum, Nimbalyst, and Kuse each published the comparison you likely found them in, so weigh their self-descriptions with that in mind. The wider field also includes chat-driven OpenClaw and Perplexity's Computer, which the same guides cover (Kilo's guide).

How to choose

Five questions separate these tools, and the table below scores each one row by row: can you read the source, can you pick the model, where does it run, where does your configuration live, and what does a start cost. If those are the answers you want, Kortix is built that way: start from the homepage, or go straight to the full head-to-head with Claude Cowork. Kortix also keeps deeper head-to-heads of this field on its blog: Kortix vs OpenWork, Kortix vs Gumloop, and Kortix vs Eigent and Kuse.