Yes. Images are supported and can be used for visual analysis, examples, comparisons, or reference.

Yes. You can upload spreadsheets, including budgets, forecasts, schedules, pricing models, or planning worksheets. These are commonly used in business coaching, event planning, and financial discussions.

Yes. Video and audio can be uploaded for review, breakdown, feedback, and coaching.

Session files are tied to one session only. Persistent files live in your profile and are always available unless you delete them.

Yes. That’s the entire point of persistent files. Upload once, use many times.

No. Uploading reports works extremely well and avoids integration headaches.

Not yet for most users. Phase 1 supports uploads. Phase 2 introduces inbound APIs for automated data feeds.

Phase 1 uses uploads. Phase 2 supports inbound APIs where external systems push data directly into sessions.

Those applications give you metrics. Replicator Coach turns those metrics into coaching context, patterns, explanations, and next-step guidance that evolves over time.

Yes. You are not locked into a single coach.

A coach persona represents a specific style of coaching, expertise, and mindset. Think skill-specific or philosophy-specific guidance.

Yes. Coaches can review sessions, add written feedback, and even voice feedback. This is a premium feature.

No. Metrics are dynamic and defined per skill.

Scores are generated by the AI coach based on:

• Uploaded files

• Session inputs

• Historical sessions

• Skill-specific evaluation criteria

A Report Card is a structured breakdown of your performance by metric, including scores and commentary.

Yes. They evolve as your inputs, skills, and sessions change.

This button tells the AI to only look at your previous sessions, not generic advice. It answers questions like:

• What did I do differently last time?

• Where have I improved?

• What mistakes do I keep repeating?

Because it turns the system into a mirror, not just an advisor.

Strategy, pricing, budgets, planning, marketing, sales, leadership, speaking, operations, and decision-making.

Yes. Business plans, budgets, spreadsheets, and planning docs are commonly used.

The platform is designed to complement human coaches and make them more scalable and consistent, not eliminate them.

No. It works only from provided inputs and defined coaching frameworks.

Yes. Follow-up chat is part of every session.

Yes. Sessions are saved and searchable.

Yes. Sessions can be filtered by coach, date, or skill.

Yes. Audio playback is supported when enabled. This is based on your actual coach’s voice and personality.

Continuity, context, skill-specific metrics, coach personas, and evolving feedback over time.

People who want improvement over time, not one-off answers.

No. It removes repetition and low-value work so you can focus on higher-impact coaching.

Both. The AI handles first-pass analysis and consistency. You step in where nuance, judgment, or human context matters.

A coach persona defines your coaching style, tone, expertise, and evaluation framework. It’s not just a name, it’s how you think.

Yes. Many coaches create different personas for different skills, levels, or coaching philosophies. A baseball coach for example might have one persona for batting, one for pitching, and one for fielding. A golf coach might have one for driving, one for obstacles, and one for putting. The system is designed to be flexible.

Yes. You can see a client’s full coaching history. Then, you can add your own commentary, including vocal feedbacck, that become part of the user’s history.

No. It is constrained to your defined approach. Guardrails are part of persona and skill setup to prevent it from answering certain things (or forcing it to answer in a certain way based on your expertise).

One coach can now serve far more clients without sacrificing quality. Subscription access, hybrid coaching, group programs, sponsorhsips, tiered reviews and other premium features can all generate new revenue opportunities. Your coaching expertise becomes an asset, not just your time.

Absolutely. This enhances live coaching rather than replacing it.

There is no single monolithic “prompt file”. Instead, the final prompt is composed step-by-step from multiple sources and then sent to OpenAI as a single request.

At a high level, the flow is:

1. Identify which OpenAI Assistant ID to use

2. Gather contextual text inputs

3. Append user-provided content

4. Attach files (both session uploads and persistent files)

5. Send everything to OpenAI in one request

Order and role determine influence, not numeric weighting.

Below is the effective hierarchy as implemented today.

1️⃣ OpenAI Assistant (highest structural authority)

The selected Assistant ID is loaded first. This assistant already has:

• Its own system instructions

• Its own training context

• Any files previously uploaded directly to the assistant in OpenAI

This assistant defines the baseline behavior of the model.

Important point:
If the assistant instructions say “You are a strict golf coach,” everything else is interpreted through that lens.

This is the closest thing you currently have to a “hard weight.”

2️⃣ Coach personality (admin settings)

Next, the coach personality text from admin settings is injected into the prompt.

This typically includes:

• Tone

• Coaching style

• Perspective

• Domain framing

This content is appended as instructional text, usually as part of the system or pre-user message.

Impact:

• Strong influence on voice and framing

• Cannot override assistant instructions

• Can be diluted if very long or vague

3️⃣ Metrics definitions (skill-specific schema)

Metrics definitions are injected after personality and before user input.

Examples:

• Batting timing definitions

• Golf swing metrics

• Speaking cadence, clarity, confidence

These act as reference constraints, not commands.

Impact:

• The model uses these as evaluation criteria

• They guide analysis but do not force outcomes

• If they conflict with assistant instructions, assistant wins

4️⃣ Session prompt (user intent framing)

This is the explicit session instruction such as:

• “Analyze my swing”

• “Evaluate this speech”

• “Give me feedback focusing on timing”

This is usually appended just before user content.

Impact:

• Strong directional signal

• Overrides general assistant tendencies

• Does NOT override safety or system-level instructions

This is the most powerful user-controlled input.

5️⃣ User uploads (session files)

Files uploaded during the session are attached to the request.

This includes:

• Video

• Audio

• PDFs

• CSVs

• Images

OpenAI treats these as primary evidence, not instructions.

Impact:

• High factual influence

• Low stylistic influence

• The model reasons from these, not because of them

6️⃣ Persistent uploaded files (profile tab)

Persistent files are also attached, but typically after session uploads.

These are treated as:

• Background knowledge

• Long-term context

• Reference material

Impact:

• Weaker than session uploads

• Can be ignored if not obviously relevant

• Competes with token limits

7️⃣ Follow-up chat history (if applicable)

In follow-up chats, previous messages are appended as conversation history.

Impact:

• Medium influence

• Decays as conversation grows

• Can create drift if not periodically anchored