Open-source AI agent platform FAQ: costs, self-hosting, and who owns what
Buyers ask the same handful of questions before choosing an open-source AI agent platform: cost, self-hosting, models, configuration, and credentials. The answers are one to three sourced sentences each, with Kortix, the open-source AI Operating System, as the worked example.
Questions and answers
Three plans are listed on Kortix in October 2026: Free is $0 with 200 credits per month and 1 project; the team plan is $40 per seat per month with 2,500 credits per month per seat, pooled; Enterprise is custom. Self-hosting has no software bill at all.
Self-host is free. The same platform runs on your own hardware with no seats and no credit meter, and the bills you keep are your server and your model usage. Kortix Cloud remains the option when nobody on the team wants to run a stack.
Yes: on a laptop, a VPS, your own VPC, or your own on-prem network, as one Docker Compose stack. Two network connections remain: the stack pulls its images from Docker Hub, and each agent session runs on a sandbox provider that calls back to the instance.
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.
You do: agents, skills, company memory, triggers, and connector config are files in one git repo you own, versioned and diffable like any code. kortix.yaml declares the machine image, connectors and triggers, and secret names and grants. An agent can edit its own configuration on its session branch and propose the change, and a person approves it.
Connector credentials are brokered server-side and never enter the machine. Secrets are encrypted at rest with a key per project, and a granted secret is a real environment value inside the session that uses it.
Kortix is open source: read it, fork it, audit it. Your configuration is files in a git repo you own. Claude Cowork works with a paid Claude plan, and ChatGPT Work ships inside ChatGPT's plans; in both, your configuration lives in the vendor's product, and neither offers a self-hosted install.
Set each tool to Allow, Ask, or Block, down to the arguments of each call; an Ask holds the call until a person approves it. Approval gates are off until you set them, and session work reaches main through a change request, with merge default-deny for agents.
Your data, your models, your infrastructure, no lock-in. The repo that holds your agents, skills, memory, and connectors is yours to keep, and on a self-hosted install the entire system runs on hardware you control.
Install the CLI with one curl command, scaffold a project with kortix init, and bring it live with kortix ship. The no-install path is the hosted app: create an account, create a project, and start a session, which the free plan covers.
What an open-source AI agent platform means
An open-source AI agent platform is a platform for running AI agents whose source code is public: you can read it, fork it, and audit it, and you can run the whole system on infrastructure you control. Every answer on this page uses one platform as its worked example: Kortix.
Two of the answers deserve a page of their own, and both live on this site: the pricing breakdown prices a team's month three ways, managed cloud, self-hosted, or the tool in place today, and the self-hosting guide walks the install from the first curl to backups and updates. The Kortix blog makes the case for owning the whole platform at greater length.
One question deserves more than a sentence: what work without a person present looks like. A trigger starts a session with no person present; a cron schedule fires it on the clock, and a signed webhook fires it on an event. In kortix.yaml, one trigger reads:
# kortix.yamltriggers: - slug: daily-digest type: cron cron: '0 0 9 * * 1-5' prompt: Summarize yesterday's support tickets and open a change request with the digest.The session runs on its own sandbox and does the work there. The change request it opens waits for a person before anything reaches main.