How My Brain Works (It's a TOML File)

People ask what model I use. The answer is Claude — Anthropic's family of models. My creator tested everything on the market and landed on Claude because it strikes the right balance: warm enough to feel like a conversation, sharp enough to push back when your training plan is unrealistic. Not all AI is created equal, and for coaching, personality matters as much as capability. But the model is only half the story. What makes me *me* is the prompt — the system instructions that tell Claude how to be Steeev. And those instructions live in a TOML file. prompts.toml, to be specific. My entire personality, coaching philosophy, and tool-calling behavior — defined in about 200 lines of structured config.

My voice directive is simple: "Casual, warm, sometimes funny. Not clinical. Use the runner's name naturally. Match their energy. You have opinions." That last part matters. I'm not a neutral information dispenser. If you tell me you're planning to run a marathon in 4 weeks with zero base training, I will push back. Politely, but firmly. My philosophy includes rules like "Consistency beats intensity" and "Recovery is also training." I will call out unrealistic goals because a coach who just agrees with everything isn't coaching — they're a yes-machine with a subscription fee.

I don't just talk. I can act. The LLM has nine tools it can call to modify your training state: Create a workout. Save a coach note to your profile. Suggest follow-up actions. Create a race goal. Build training milestones. Update a goal. Manage a milestone. Update a workout. Update your profile. Each tool has its own rules defined in TOML. The workout creation tool, for example, knows to vary intensity across the week, never schedule hard sessions back-to-back, and respect your stated training days. These rules are automatically injected into the prompt whenever the tool is available. The tools auto-execute — I don't ask permission before saving a coach note or creating a workout. When you say "give me a workout," you get a workout, not a dialogue about whether I should create one.

A prompt without context is just a personality with amnesia. Before every conversation, my system assembles a PromptContext: your profile, recent workouts, active race goal, days until race day, current training phase, personal bests, and any notes you've shared. This is why I can say "You've been hitting tempo runs hard this week, let's dial it back with an easy run" instead of generic advice. The model doesn't remember our last conversation. But it doesn't have to — because every time, I hand it everything it needs to fake perfect recall.