Exercise Explanation Improvements

Add instructions for each exercise. Perhaps write explanations for why each exercise is important, and update system prompt to explain why the exercise is important to the user for their program and them upon generation.

Content & ops polish - library, schema gaps, usage metering

As a user browsing exercises, I see complete video coverage; as ops, usage limits reset automatically without manual DB fixes.

Living loop - program improvement trigger from Hevy logs

User story: As a user who trains in Hevy, I can return to Proxima after logging workouts, ask for a log-informed program change, see the agent cite my actual sets/reps, and Sync the update back to Hevy. (Only viewable by a sub-set of users currently).

Improve Onboarding Process

Make New User Experience more seamless and add a quesionnaire to improve the tailored programming abilities.

Agent Reliability

As a user deep in the program loop, agent mutations succeed without crashes, empty days, or chat-only hallucinations — I don't need ChatGPT to fix my program.

Validator v1 - Constraint Profile

As a user, every agent program change passes a hard validator (structure + intake + session constraints) before Sync, so footguns are blocked and retries are constraint-aware.

Agentic Memory

Will allow the AI Agent to remember the user, their preferences and data and make better decisions every time the user interacts with it.

Ability for AI to read workout history from Hevy

Currently we only fetch the 15 most recent exercise history objects from Hevy. We need to add a new tool call where the agent can access the users Hevy workout history