Self-host an open source Perplexity Computer alternative

What self-hosting buys you

Self-hosting an open source Perplexity Computer alternative runs the whole system on hardware you control, and four things change. Your data and credentials stay on your side: the repo, the memory files and the brokered connector credentials all sit on machines you own. You bring your own model keys, and Kortix is open source, so you can read, fork and audit the code you run. There is no per-seat charge on a self-hosted instance: self-host is free, and you pay for your own compute and model usage. The trade is operational: you own uptime, upgrades and isolation, which means patching the host, watching the Docker Compose stack and backing up the data yourself.

The four decisions

Self-hosting Kortix comes down to four decisions: which project to run, which host to prepare, which models to connect, and which rules to set for agents. The commands for the install and start steps are in the code panels on this page.

1. Pick the project

Perplexity Computer is a hosted, closed product: it runs only in Perplexity's cloud, and its local Portable Computer build runs a runtime you cannot audit (Vellum's breakdown). Self-hosting therefore starts with software that publishes its source, and the pick here is Kortix, the open-source AI Operating System. Kortix self-hosts on a laptop, a VPS, your VPC or on-prem, and one git repo holds the whole system.

Other open source projects run on hardware you control, with narrower scope. OpenClaw is a personal assistant that runs a Gateway on your own computer and answers in the chat channels you already use. Eigent is a cowork desktop with local or self-hosted deployment and a review step before you accept a result. AnythingLLM runs entirely on your device for chat over your own documents. Vellum runs on your Mac or in Vellum's cloud. The alternatives page scores the wider field.

2. Prepare a host

Any machine you control works: a laptop to try it, a VPS for a team, a VPC or an on-prem server for regulated work. Kortix runs as one Docker Compose stack that holds the frontend, the API, the LLM gateway, and the Supabase distribution; agent sessions run on a separate sandbox provider, not on this stack (Kortix self-hosting docs). Three commands take a bare Linux box to a running system: install the CLI, initialize the instance, then start the stack. The first run asks six things and no model key, GitHub connects in the dashboard under Settings → Git, and the stack has no Redis and no separate worker. The instance needs the internet, because kortix self-host start pulls images from docker.io and the sandbox must be able to call back.

3. Connect your models

Give each agent the model it should run on. Any model provider with your own keys. Or the ChatGPT plan you already pay for. Or sign in with your OpenCode Console account for OpenCode Zen and Go. On a self-hosted instance you connect your own LLM key in the model picker, and self-hosted instances use your own key by default. Changing the model later is a config edit in the repo you already own.

4. Harden what agents may touch

Lock the system down before real work runs. Connector credentials are brokered server-side and never enter the machine, and secrets are encrypted at rest with a key per project. Set every tool to allow, ask or block, down to the arguments of a single call. Approval gates you set. Off until you set them. One isolated sandbox per session. Each session has its own isolated machine and branch. Session work reaches main through a change request. Merge is default-deny for agents, so nothing lands without a review.

Where to run it

The same install runs on a laptop, a VPS, your VPC or your own rack, and Kortix Cloud is a fifth option when you want no operations of your own. Pick the host by what you can maintain.

Why Kortix is the self-hosted pick

Kortix is the self-hosted pick because the whole system is one repo and one command surface. One git repo holds the agents, the skills, the company memory, the connector config and the triggers, so the whole system is text you can read, diff and roll back. The connector surface is 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP. Run it on Kortix Cloud, in your VPC, or on your own on-prem network; self-host is free. OpenCode is the harness, and the configuration stays files you own. Read the source at Kortix on GitHub. For a Kortix-against-Perplexity breakdown, read how Kortix compares with Perplexity Computer and the review of Perplexity Computer; for the field-wide view, see the Kortix blog.