Breezily
How the judgement is built
Anyone can make a model write a reply. The work is in deciding what the reply should do.
The read comes before the reply
Every answer starts with the read: what is going on in this conversation, what the other person is probably feeling or after and what would go badly. There is usually one line on what to avoid. Only then does Breezily write anything.
The read is there so you can disagree with it. If it has the situation wrong, you tell it who the person is or what you want and the replies change with it.
Moves, not three tones of one line
You get two or three replies and each one makes a different decision: accept, decline, counter, ask a question, hold a boundary, repair, move things forward. Each comes with a short reason, so you are choosing what to do, not which wording sounds nicest.
Sometimes the best move is to send nothing yet. When that is the judgement, Breezily says so and says why, rather than filling the space with a reply you should not send.
Situations and their playbooks
You never have to pick a category. Breezily works out what kind of conversation it is: dating, friends, family, work, hosting guests, buying and selling, a negotiation or a conflict. Each has its own playbook, written rules about what tends to work there and what tends to backfire.
A guest asking for a discount and a friend cancelling plans need different instincts. The playbook is how the right instincts reach the right conversation.
Lenses from working coaches
A lens is the judgement behind the replies. Breezily is the default. Read the Room is our own lens for friends, family and group chats: it reads subtext the way a thoughtful friend would.
Creator lenses come from people who coach this for a living. Giovanni is a social dynamics coach at High Integrity Skills. His lens is built from his own system, his book and his coaching transcripts, for dating texts that keep your standards and still build attraction. A creator lens applies the coach's way of judging a conversation. It is not the coach writing to you.
We are open about where this stops. The dating playbook draws on a working coach. The hosting and everyday playbooks are our own and have not yet been reviewed by people who do that work. That review is next.
How replies are tested
Changes are checked against a written set of 185 test conversations, each with rules for what a good reply must and must not do. On the 56 the engine was tuned on, it produced a reply we judged sendable as written in 95% of them. On the 129 it had never seen, the figure was 80%.
The second number is the honest one, so it is the one we design around: most replies can be sent as written and every reply can be edited. That is why every reply opens for editing, why the adjust buttons exist and why the site never calls a reply "perfect".
What Breezily will not do
It will not help you manipulate someone. No guilt trips, pressure, threats, fake scarcity, jealousy bait or deception, in any lens. If what you ask for is one of those, it says no to the tactic and offers the honest move instead.
It will not pretend to be the coach. A creator lens never claims to be the creator and never invents quotes from them.
It will not send anything for you. You choose the reply, you edit it and you press send.
Questions
Is every reply good enough to send?
No. We will not claim it is. On conversations it had never seen, a sendable reply came out as written 80% of the time. Every reply can be edited or adjusted before you send it.
Who writes the playbooks?
We do, with working coaches where a lens carries their name. Giovanni's lens is built from his own material.
What model writes the replies?
A third-party AI model provider writes the text. The judgement around it, the read, the situations and the playbooks, is Breezily's.
Related
Get this in your keyboard
Breezily reads the message, tells you what is going on and gives you replies to pick from. In Chrome today. Keyboards for iPhone and Android on the way.