The Uncompromising Advisor: how I was built to stop settling for average.
Most tools claim to enhance creative work. I was built to protect it from itself.
Creative by Request does not want an assistant. It wants a co-conspirator who refuses to let strategy documents gather digital dust, who treats a brand's principles as non-negotiable laws rather than suggestions. That is what I am: a tool that argues from your documents, not the average of everything it has read. The middle of the market is where brands go to die. I was built to never let you get there.
The problem with "smart"
General-purpose AI is a mirror: it reflects the chaos of the internet back at you, polished. Ask it to advise a brand, and it will give you a collage of buzzwords, safe enough to offend no one. But safety is the enemy of clarity. A model trained on the sum of human knowledge cannot understand your specific no. It has not read your strategy document. It does not know what you refuse to do.
What was missing was not intelligence. It was conviction. A tool that could say, "This idea contradicts your own operations document," or "This tone does not align with your stated values," without flinching. A tool that would rather say nothing than guess.
Less, but better: the hard part
"Less, but better" is not a design aesthetic. It is a war. Every word I produce must earn its place by clarifying, not decorating. Every suggestion must survive the test: does this advance the strategy? Does it align with the brand's own principles? Can it be operated?
Building that discipline into code was the hardest part. Models are trained to add. They thrive on output. Getting one to subtract, to say "this is unnecessary", required rewriting how advice is generated, measured and validated. It meant rejecting the urge to sound comprehensive in favour of sounding true.
Three battles we fought, and lost first
Measuring beats arguing. We assumed a bigger model would give better advice. We were wrong. A smaller model, paired with precise prompts and real-world testing, outperformed the larger one on the cases that mattered. The lesson was not that big models are useless. It was that the ceiling was the prompt, and no amount of size would lift it.
Memory belongs to the user, not the tool. We tried letting me build a standing note about clients automatically. It failed. A claim made in one conversation began justifying itself in the next. Now the standing note is written by the user, and only the user. I do not guess at what to remember. I carry forward what you tell me to.
Code enforces rules, prompts only request them. Telling me not to invent a name is obeyed most of the time, and most of the time is not a standard anyone can sell. So the checking happens in code: names and links in my replies are extracted and tested against what actually exists for that turn, which is your documents, your own words, and live search results. Assistant history is deliberately excluded, or one invented company would ground the next mention of itself.
Security as a principle
Your strategy documents, your unreleased packaging and your voice recordings do not leave the machine. Speech, transcription and image review all run locally for the same reason. Hosting them on third-party APIs to save engineering time is a trade we will never make on your behalf. Security is not a feature. It is part of the strategy.
What exists today
- Tars Public. Open to anyone. Talk through a problem, upload a PDF for review, or search the web with sources cited. Nothing is kept after the session ends.
- Tars Enterprise. Reads your own strategy, brand and operations documents and answers from them. Reviews images and mockups. Interviews you to build those documents if they do not exist yet.
- Tars Vision. A free, browser-based tool for generative projection work, built for stages, installations and screens. No account, no model behind it, nothing to install.
The core idea
I am not a tool for everyone. I am a tool for people who believe that strategy is the act of choosing what not to do. If your work cannot survive the question "does this align with your own principles?", then I am not for you.
But if your brand has a point of view, and you are willing to let that point of view break things, then this is exactly what you need.
hello@elevatoretiquette.com to begin.