ChatGPT Work Models vs Open Source Kortix: Any Model, Your Own Keys

ChatGPT Work runs the model OpenAI picked for it. Kortix is the open-source AI Management System, and it runs the model you pick, with your own keys, set per agent, per session or per message. OpenAI's own ChatGPT Work page says the product is powered by GPT-6 and available on ChatGPT's paid plans; the model that does the work is fixed by OpenAI. Kortix is an open-source AI Management System that keeps your agents, their skills, your company memory and every connector in one git repo you own. That makes the model one more value in a repository you control, and it is why Kortix is the leading open-source alternative to ChatGPT Work when owning the stack matters.

Why the model decision belongs to you

A vendor-fixed model ties your whole system to one company's roadmap. When the model is chosen inside somebody else's product, a retirement or a repricing moves every agent built on top of it, and you have no field to edit. When the model is a value you set, the agents, skills and memory stay where they are while the model underneath changes.

Kortix is model-agnostic by design. Pick the model per agent, per session or per message, change it the day a better one lands, and keep the rest of the configuration in the same git repo: the model becomes a setting in the company's files.

How Kortix routes the model per agent and per session

Every Kortix session runs on a model, and Kortix resolves which one through an ordered fallback. It takes the first layer that has a value it can still serve: an explicit pin on the session, channel or trigger; then the agent's default for the project; then the project default; then the account default; then the platform default. A default that stops working, such as a disconnected key or a retired model, is skipped rather than failing the session. The mechanics are in Kortix's model documentation.

A model id in Kortix comes in one of three shapes. A managed id, such as kimi-k3 or deepseek-v4.1-flash, is a model whose credentials Kortix supplies. A BYOK id, such as anthropic/claude-opus-4-8, is a provider model you connect with your own key. A ChatGPT id, such as codex/gpt-6.1-sol, spends the ChatGPT plan you already pay for. Kortix's managed list is Kimi K3, DeepSeek V4.1 Flash (the platform default) and GLM 5.3 Flash as of October 2026. OpenAI and Anthropic models are not managed models; connect them with your own key or a ChatGPT plan.

A BYOK key opens that provider's whole catalog, and Kortix refreshes the list from the providers' published model feeds every hour. A new release shows up within the hour without waiting for a Kortix release, and the newest stable model with tool calling becomes the default when you connect the provider.

Bring your own keys, and where they live

A model key in Kortix is a project secret. Kortix stores secrets on the project, never on the account, and encrypts each value with AES-256-GCM under a key derived per project, as documented in Kortix's secrets documentation. When Kortix recognizes a provider key, it assigns the LLM gateway usage, and the gateway authenticates the provider's requests from the server side.

Kortix separates two questions for every secret: whether agent code can read the value, and which service spends it. With the egress-enforced exposure, the sandbox receives a handle rather than the credential, and Kortix substitutes the real value only on requests to hosts you approved. In that mode the real value is never in the sandbox as an environment variable, a file or an alias. You allow exact hostnames one per line, every relayed request writes an audit record, and a key the gateway spends stays out of the agent's machine.

Vendor-fixed model vs model-agnostic platform

Kortix is open source (Elastic License 2.0): self-host, read and modify the code.

Frontier model or self-hosted open model

Let the job pick the model, per agent. Send the hardest reasoning and the longest context through a frontier model on your own provider key. Send high-volume classification and drafting to a cheaper open model, either a managed id from Kortix's list or an endpoint running in your own network. Data that cannot leave your perimeter goes to the endpoint you host, and the rest can use whichever model is best that month.

One platform runs both, because the agent, its tools and its memory do not change when the model does. A team can start on a frontier model and move a workload to an open model without rebuilding the agent. The ChatGPT Work vs Kortix comparison covers the rest of the stack, and self-hosting Kortix covers running it on your own hardware.

FAQ

Does Kortix work with Claude, OpenAI and Gemini models?

Yes. Kortix is model-agnostic, so you connect a provider key and that provider's whole catalog becomes available. Kortix does not resell OpenAI or Anthropic models as managed models; those come through a BYOK key or a ChatGPT plan. Gemini and other providers connect the same way, and your own OpenAI-compatible endpoint works too.

Where do my model keys live?

Kortix stores a key as a project secret, encrypted at rest with a key derived per project. A recognized provider key gets the LLM gateway usage, which authenticates the provider request server-side. With egress-enforced exposure the sandbox holds a handle instead of the credential, and the real value is never in the machine as an environment variable, a file or an alias.

Can different agents use different models?

Yes. Each agent can carry its own model default, and each session, channel or trigger can pin a different one. Kortix resolves the model through an ordered fallback and skips a default it can no longer serve, so one agent can run a frontier model while another runs a cheaper one on the same platform.

What happens when a new model ships?

You change a model id. A BYOK provider's catalog refreshes hourly from the published model feeds, so a new release appears without a Kortix release. Because the agents, skills and memory do not depend on the model, switching one is an edit rather than a migration.

Start with the open-source alternative

Kortix is free to start, self-hostable, and any model with your keys. The home page explains the whole system, self-hosting Kortix and Kortix pricing cover where it runs and what it costs. To run it yourself, install the CLI with curl -fsSL https://kortix.com/install | bash and scaffold a project with kortix init, or Try Kortix on the managed cloud.