Models
The AI model you choose, not the one you are given
Pick from 15 providers, or plug in a model hosted on your side. Keys are managed in one place for the whole organization.
15 providers, or your own
The main model providers are ready to plug in. For a model hosted on your side, two compatible endpoints are enough.
Gemini is used through OpenRouter or an OpenAI-compatible gateway.
Add a provider
One key for the whole organization, chosen models
An administrator adds a provider's key once, and ticks the models teams will be able to use. Nobody else has a key to handle.
- Two steps: provider, name and API key, then “Choose models”. Nothing is created before “Add”.
- “Test” checks that the key works.
- Changing the key, reviewing the models or deleting the provider is done from its row.
Default model
One model for everyone, another where it matters
The organization has a default model. Every agent inherits it, unless someone picks another one in its Settings. To try something, you switch models for a single run.
- “Set Claude Sonnet 5.5 as default”: one click, and the agents that inherit switch models. Without a default, a warning says so.
- In the agent, “Model used” keeps “Inherit” or sets a specific model, with its temperature and reasoning level.
- “Run with options…” tries another model on a single run, without touching the agent.
Personal keys
Your personal key, if the organization allows it
Some members already have their own key or a subscription. An administrator decides in one click whether they can use it in Appstrate.
- The “Members can use their personal key or personal subscription” setting is turned on for the whole organization.
- Each person adds their keys in their Preferences, under “My model keys”.
- A schedule never uses a personal key: an automatic run does not depend on a member's account.
Model hosted on your side
A model on your side, plugged in like the others
Ollama, vLLM or LiteLLM plug in through an OpenAI-compatible or Anthropic-compatible endpoint. Agents use it like any other model.
- Enter the base URL and the API type, then “Detect models”.
- Local addresses are blocked by default: the server operator has to allow internal hosts.
- With a local model, exchanges with the model stay on your network.
What it changes
For your teams
- The model is already chosen: a new agent inherits the default model.
- To try something, “Run with options…” switches models for a single run.
- If the organization allows it, you use your own key or subscription.
For the organization
- One organization key serves every agent: no keys to hand out.
- Switching providers does not mean rebuilding agents: the ones that inherit follow the new default.
- No dependency on a single provider, and a local model is possible for sensitive data.
Budgets for models, per team and per agent
Set a spending cap in dollars for the whole organization, a space, a person or an agent, separately for the chat and for agents. A budget that is reached blocks what it covers, and the refusal names it.
- An alert to the owner at 80%, with an adjustable threshold.
- Per day, week or month, with exceptions per person.
- A Claude or ChatGPT subscription counts in dollar equivalent.
- In the open-source core: no paid module needed.
Frequently asked questions
Is Gemini available?
Who pays for model calls?
What does the “Test” button check?
Can I use a Claude or ChatGPT subscription?
Let's see what it would look like on your side.
Try it on your own in a few minutes, or take thirty minutes with us to scope a first use case.
Let's talk about your case
Thirty minutes with us to scope a first use case and its gain.
Read the code
Apache 2.0. Your technical team can read it all, install it all, keep it all.