Copilot

The copilot is a chat inside the app that operates Aseptic for you: it starts and stops scenarios, calls endpoints, creates mocks and explains why something won’t start. You ask it “launch the scenario and tell me why orders-bff fails” and it does it.

It uses exactly the same tools as the MCP server (see MCP and copilot), so it can do the same as an external agent. It’s a paid capability (chat).

Provider and API key

The copilot talks to the provider you choose, always with your own account: you pay for the consumption and Aseptic does not sit in the middle. There are three ways to pay for it:

With your Claude subscription — no API key, no credit. It is the most direct option if you already pay for Claude Pro or Max. The first time you use it, Aseptic downloads Claude Code (about 300 MB) to its own folder; after that, never again.
With your own API key from the provider: Anthropic, OpenAI, Google AI Studio, xAI (Grok), DeepSeek, Mistral AI, Groq, NVIDIA and OpenRouter (a single key that reaches the models of nearly every vendor).
With a local model, no key and without leaving your machine: Ollama and LM Studio.

On the sign-up screen each provider states where its key is obtained and which prefix it starts with, so there is nothing to look up elsewhere.

The key is stored encrypted with the system’s key store, never in cleartext. If the system’s encryption is not available, Aseptic does not store it.

Models

The model list is discovered live against the provider’s API, it does not come compiled into Aseptic: as soon as the provider publishes a new model it shows up in the selector, with no app update. Aseptic ships a fallback list so the selector is not empty before the first refresh.

Destructive actions

Before a destructive action (stop, clean, overwrite…), the copilot asks for confirmation. You authorize or deny at the moment. The list of what’s considered destructive is the same one the MCP marks.

The Auto toggle, next to the input box, changes that: while it’s on, the copilot authorizes each destructive action itself and doesn’t stop the conversation to ask. It ships off and the preference is remembered between sessions. Handy for a long run of work on a throwaway scenario; with a catalog you care about, leave it off.

Usage

You type in natural language; Enter sends and Shift+Enter makes a line break. You can stop a response in progress, start a new conversation from the header menu and choose the model.

Saved conversations

Each conversation is saved on your machine after every answer, so closing the app does not lose it. The copilot opens on the conversation list: the New conversation card at the top and the previous ones below; click one to resume it as it was —the copilot remembers what was said—. The text box appears once you are inside one, and the arrow at the top takes you back to the list without stopping an answer in progress. A long conversation opens at the end, with a button to show the earlier questions. Each one is titled by the copilot itself in its first reply; from the row’s menu you can rename it —from then on the title is yours and the copilot no longer changes it— or delete it. The 50 most recent are kept.

A conversation is a single one for every model. You can switch provider or model midway, or resume it with a different one from the one that started it, and whoever takes over carries on with everything said. If the one answering is the same one that spoke last, it gets the full conversation back; if it is another, it receives what was said and a summary of each action run (if it needs the full result of one, it asks for it again).