Cosine is an AI software engineering agent designed to integrate seamlessly into your team's workflow.

It autonomously retrieves data, plans solutions, writes and tests code, and collaborates in real time or asynchronously – freeing your team to focus on innovation instead of repetitive tasks.

Cosine uses a seat-based subscription model with a monthly token pool per seat.

Hobby — $20 per seat / month. Each seat includes 5M tokens per month. Additional seats each add another 5M tokens to your team's monthly pool.

Professional — $200 per seat / month. Each seat includes 60M tokens per month. Additional seats each add another 60M tokens to your team's monthly pool.

Token top-ups are available on both plans if you need more capacity mid-month.

Token credits are the currency Cosine uses to run agents. Every time Cosine reads code, plans a solution, or writes output, it consumes tokens from your team's monthly pool.

Hobby seats include 5M tokens per month and Professional seats include 60M tokens per month. Unused tokens do not roll over. If you need more, you can purchase top-ups at any time from your billing settings.

When your team's token balance is exhausted, Cosine will pause new agent inference until the balance is replenished.

You can purchase a token top-up instantly from your billing settings to resume work, or wait for your monthly pool to renew at the start of the next billing cycle.

Seats refer to individual user licenses within a team plan. One seat equals one user. Teams can add multiple seats to enable collaboration across members within the same workspace.

Each seat also contributes to your team's monthly token pool — so adding seats increases both the number of users and the total tokens available.

A task is an execution of the Cosine agent. You can retry a task or continue the conversation within it as many times as needed. Each task consumes tokens from your team's monthly pool.

Projects are the code repositories where tasks are executed. They provide context for the agent and organise related work.

Cosine is not trained on customer data or private code repositories. Its capabilities are based on training from publicly available open-source repositories collected outside the platform.

We also maintain enterprise agreements with all AI vendors to ensure that no data sent to them is used for training purposes.

Users retain full ownership of their projects and code at all times.