Skip to content

Model choice

Any AI model you like, not a shortlist we picked for you

Steptree is not a wrapper around one vendor’s model. Over two hundred models from more than twenty labs are in the picker: frontier and open-weight, large, small and specialised. Every plan reaches all of them, and you change your mind with a dropdown. What you are buying is your own procedures written down, and read by the AI you choose. The model is just the engine, and you can swap it.

Why it matters to you

What that means when you are actually using it

All four are true on every plan, including the free one, from your first minute.
Your plan does not decide which models you get
Every plan reaches every model in the catalogue, frontier ones included and the free plan among them. What a plan buys is how much you can run, not which engines you are allowed to run it on. A dearer model simply spends credits faster.
Switching is a dropdown, not a migration
Each agent carries its own model, and in the desktop app each conversation does too. Change it and the next run uses it — nothing is recompiled and nothing is redeployed, because a procedure describes the work rather than the model doing it.
New models appear without waiting for us
The list refreshes every day, so a model released this week shows up in your picker this week, priced and ready. You do not wait for a Steptree release to use it.
A model retirement does not interrupt your agents
When a lab retires a model, the work it was doing is a procedure you own, not a prompt tuned to one vendor's quirks. Pick another model, and the agent carries on doing the same job.

The catalogue

Open-weight through frontier, in one list

One picker, one price list, one bill, whichever lab you choose from. The list refreshes daily as labs publish and retire models, so it is never the one we shipped.
  • OpenAI
  • Anthropic
  • Google
  • Meta
  • Mistral
  • DeepSeek
  • xAI
  • Alibaba (Qwen)
  • Moonshot
  • MiniMax
  • Z.ai
  • Amazon
  • NVIDIA
  • Cohere
  • and more, added as they publish

Picking well matters more than picking once. A cheap model is usually right for reading a document or filing a ticket, and a frontier one earns its cost on a judgement call. Because every plan reaches both, you can put the expensive model on the one agent that needs it instead of paying for it everywhere — see how credits work.

Or none of ours

Run Steptree AI's model on your own hardware

On Team and above, Steptree AI in the Steptree App can use a model on your own network instead of a hosted one. Any OpenAI-compatible server works, so Ollama, vLLM, LM Studio and llama.cpp all do. It sits in the same picker as the hosted models, and the address is checked as you type it. That conversation then goes to your hardware instead of to a hosted model. The app still signs in to Steptree over the internet, and writing an SOP or compiling a skill always uses a hosted model.

The other kind of lock-in

Your procedures, in a format anything can read

Being free to change model is worth little if the documentation is trapped.
An open file format
Any SOP compiles to SKILL.md — YAML at the top, markdown below, the format other AI tools already read. Nothing about it is ours.
Downloads on every plan
A compiled skill downloads as a file on any plan, with no request to raise. The free plan compiles 2 skills; paid plans have no limit.
Readable without us
A downloaded SKILL.md file is a procedure a person can read in any text editor, with Steptree or without it.

Questions

The things people ask about models

Which models can I actually use?
Over two hundred, from more than twenty labs. OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, xAI, Alibaba's Qwen and Amazon among them. They range from open-weight models costing fractions of a cent to the frontier models released this quarter. The list is refreshed daily, so it is the current one rather than the one we last shipped.
Does a cheaper plan mean cheaper models?
No. Every plan reaches every model, the free one included — it is the credits that differ, not the catalogue. A free workspace can run a frontier model on a real process from its first minute. When its 300 monthly credits are gone, the free models keep working until the month resets.
Can I change an agent's model after it is built?
Yes, whenever you like, and the change takes effect on the agent's next run. The agent is compiled from your approved procedure, not from a model, so swapping the model does not invalidate anything you have written or approved.
Can I bring my own OpenAI or Anthropic API key?
No, deliberately. Every hosted model runs through one Steptree account on one price list. That puts every model on every plan, with nothing for you to set up and one bill at the end of the month. If what you want is the model itself under your own control, that is the next answer.
Can we run a model on our own hardware?
For Steptree AI's chat, yes, on Team and above. Point the Steptree App at any OpenAI-compatible server on your own network, such as Ollama, vLLM, LM Studio or llama.cpp, and that conversation goes to your hardware. The app still signs in to Steptree over the internet. Writing an SOP and compiling a skill always use a hosted model.
If we stop paying, what do we keep?
Nothing is deleted when a paid plan ends. The workspace moves to Free and keeps every SOP, which you can still open, edit and share, and your AI can still read over MCP. Every skill you compiled still downloads as a SKILL.md file. Each person can also download a copy of their own data from Settings, including the SOPs they created. There is no PDF or Word export.

Next

Where to go from here

  • What agents do with it

    The work a model actually carries out for you, and the limits you hold it under.

  • How credits work

    What a run costs, what the free models cost, and what happens when an allowance runs out.

  • What stays in your workspace

    What is never recorded, how workspaces are separated, and what a local model covers.

Do one real job once, and read the SOP Steptree writes

Record it on Windows or Mac. Steptree writes the SOP with the goal, the steps and the judgement calls. Approve it, share it with your team, and connect your AI. Free to start, no card.

Someone at a wall of sticky notes talking a seated team through a process.