# Migrate from Perplexity Computer to an open-source system you own

> Moving off Perplexity Computer happens one job at a time, and Kortix is the open-source AI Operating System to move it to. You bring your own model keys, set the connector rules, and put a review gate on before the first task runs.

URL: https://opensourceperplexitycomputer.com/migrate-from-perplexity-computer

## A running instance in three commands.

`terminal`

```sh
# 1 · Install the CLI
curl -fsSL https://kortix.com/install | bash
# 2 · Scaffold a project — creates kortix.yaml + your agents, skills and runtime config
kortix init
# 3 · Ship it — pushes your repo and brings the whole thing live in the cloud
kortix ship
```

Source: https://github.com/kortix-ai/suna

The install, a scaffolded project, and the first ship, exactly as the Kortix README prints the first run.

## What a team actually moves

Perplexity Computer is a closed, hosted agent that Perplexity sells inside a paid subscription. It routes about 19 hosted models for you, reaches roughly 400 managed OAuth connectors, and meters credits on top of a $200-a-month Max plan, all recorded in [Builder.io's hands-on review](https://www.builder.io/blog/perplexity-computer) (checked October 2026). The local Portable Computer runs the same closed runtime on hardware you buy, so a Perplexity subscription still applies ([Vellum's breakdown](https://www.vellum.ai/blog/official-perplexity-computer-breakdown)). None of that is a file you can export, so the move is a rebuild of the work itself. Kortix is the open-source AI Operating System, and it keeps the same pieces as text: agents, skills, company memory, connector config, and triggers live in one git repo you own.

### The four things you rebuild

1. Tasks become agents and skills: an agent is `agents/<name>.md` for behavior plus an `agents.<name>` block in `kortix.yaml` for governance.
2. Model routing becomes explicit: you pick the model per agent, per session or per message, instead of Perplexity choosing from its hosted lineup.
3. Connectors become the connector config in `kortix.yaml`, with each tool set to Allow, Ask or Block.
4. The credit meter becomes your own compute and model bills, because self-host is free.

### Start with one job

Install the CLI, scaffold a project, and ship it, which creates `kortix.yaml` plus your agents, skills, and runtime config. Point the first agent at a single job you already understand, write it as `agents/<name>.md`, and give it a scoped reach in `kortix.yaml`. Keep the rest of your Perplexity Computer work running until that one job produces a result you trust, then move the next job.

### Connect models with your own keys

Model choice in Kortix is explicit. 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 your own key is used by default, and changing the model later is a config edit in the repo you already own.

### Set the connector rules

Connect the tools the first job needs, then set each tool to Allow, Ask or Block, down to the arguments of a single call. Connector credentials are brokered server-side and never enter the machine, so the project does not hold a raw token in the sandbox. Start with the tools the first job needs and leave the rest unconnected until you need them.

### Put the review gate on

Every session runs one agent in its own isolated sandbox, on its own branch. Session work reaches `main` through a change request, and merge is default-deny for agents, so a person reads the diff before anything lands. Approval gates you set, off until you set them, hold a tool call until a person approves it, then resume the agent's turn. [Team permissions](/perplexity-computer-for-teams) add roles, groups, and an audit trail on top.

### What changes in a repo you own

In Perplexity Computer the configuration, the model routing, and the connector set live inside Perplexity's product. On Kortix they are files in one git repo, so the whole system is text you can grep, diff, and roll back. That is the difference the move is for: the system gains an owner and a history instead of a settings screen someone else controls.

### What you cannot bring with you

Perplexity's model routing is not a file, and its hosted model lineup does not transfer. Credits you have already spent stay spent, and the managed OAuth tokens do not move, so you reconnect each tool on Kortix. Those limits are why starting with one job beats a big-bang cutover. For the underlying differences, read [how Kortix compares with Perplexity Computer](/kortix-vs-perplexity-computer), [what Perplexity Computer costs](/perplexity-computer-pricing), and [the wider field of alternatives](/alternatives). To run the whole system on hardware you control, start with [self-hosting Kortix](/self-hosting).

_FAQ_

## Questions a team asks before switching

### Can I export my Perplexity Computer setup?

No. Perplexity Computer is a closed, hosted product, so there is no export of its configuration, model routing, or connectors. You rebuild the work on Kortix by describing each job as an agent and reconnecting the tools it needs.

### Do I have to self-host Kortix to leave Perplexity Computer?

No. Kortix runs on Kortix Cloud, in your VPC, or on your own on-prem network, and self-host is free. The repo, agents, and skills are the same files either way, so you can start managed and move to your own infrastructure later.

### Which job should I move first?

Pick a single job you already understand and give it to one Kortix agent, with its own sandbox and branch. Keep the rest running on Perplexity Computer until that job returns a result you trust, then move the next one.

### Do I keep control of model choice after the move?

Yes. In Kortix you name the model per agent, per session or per message, using any model provider with your own keys, or the ChatGPT plan you already pay for. Perplexity Computer routes its hosted models for you.

### What can I not bring across?

Perplexity's routing logic, its hosted model lineup, and the managed OAuth tokens do not transfer, and credits you have spent stay spent. You reconnect each tool on Kortix, where connector credentials are brokered server-side and never enter the machine.

## Move your first job to the open-source system you own.

Start on Kortix Cloud or self-host for free, and keep every agent, skill, and connector in a repo you control.

[Try Kortix](https://kortix.com) · [Self-host Kortix](https://opensourceperplexitycomputer.com/self-hosting)

Open source · Any model, your keys · Self-host, VPC, or on-prem