# AI use cases for Social Coach (GeminiLight/SocialCoach)

Repo: [GeminiLight/SocialCoach](https://github.com/GeminiLight/SocialCoach) | Use cases: 3 | Page: https://dedupedai.com/use-cases/geminilight-socialcoach/

SocialCoach is an open-source AI EQ coach that lets you rehearse hard social conversations in role-play and then reviews your own words with personalized feedback.

This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL.

- Rehearse a hard conversation on the hosted site: https://dedupedai.com/use-cases/geminilight-socialcoach/rehearse-hard-talk-on-hosted-site/
- Run Social Coach locally with your own model key: https://dedupedai.com/use-cases/geminilight-socialcoach/run-socialcoach-locally-with-own-model-key/
- Self-host a team instance with Docker Compose: https://dedupedai.com/use-cases/geminilight-socialcoach/self-host-team-instance-with-docker-compose/

## How to rehearse a hard conversation on the hosted site with Social Coach

Repo: [GeminiLight/SocialCoach](https://github.com/GeminiLight/SocialCoach) | Difficulty: Beginner | Time: 15 minutes | Basis: From the README

Practice a tense talk in role-play with no sign-up, then read feedback quoted from your own words.

**Who it is for:** An employee or friend who has to raise a difficult topic, such as asking a boss for a raise, and wants to try it out first without installing anything.

**The problem:** Hard conversations are easy to dread and hard to rehearse alone, and it is unclear what to say or why a reply fell flat.

**The outcome:** You complete at least one practice dialogue, read a review that cites your own lines, and try a different reply in the next round.

### Steps

1. Open https://socialcoach.aurax.live in a browser. No registration is needed; the site is available in Chinese and English.
2. Pick a scenario from the library, or describe your own situation, and set your role and the goal you want to reach.
3. Talk through the dialogue. The other character has its own needs and objections and may push back. Text or browser voice input both work.
4. Read the review, which quotes what you said before giving feedback, then try another reply in a new round.
5. Optional: open the 3D practice at https://socialcoach-ai.vercel.app/3d to rehearse a dinner table, elevator, or office conversation.

### Prompt to paste into your agent

```text
Using the SocialCoach site at https://socialcoach.aurax.live, help me prepare for <target conversation, e.g. asking my manager for a raise>. Run a practice dialogue, then report the review's main point and one alternative reply to try. Verify that the review quotes my own words before its evaluation.
```

### Caveat

The hosted site is a shared demo, and voice input may send audio to the browser's speech recognition provider, so avoid speaking real personal details you would not share with that service.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/geminilight-socialcoach/rehearse-hard-talk-on-hosted-site/

## How to run Social Coach locally with your own model key with Social Coach

Repo: [GeminiLight/SocialCoach](https://github.com/GeminiLight/SocialCoach) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README

Clone the repo, add an Anthropic or OpenAI compatible key, and practice on localhost:3000.

**Who it is for:** A developer or privacy-minded user who wants to run the coach on their own machine with their own model credentials.

**The problem:** The hosted site may not fit your needs, and you want the app running locally with a model provider you control.

**The outcome:** The app runs at localhost:3000 and completes a practice dialogue using your configured model.

### Steps

1. Run `git clone https://github.com/GeminiLight/SocialCoach.git` and then `cd SocialCoach/app`.
2. Run `pnpm install`, then `cp .env.example .env.local`.
3. Edit .env.local: set LLM_PROVIDER to anthropic or openai, add LLM_API_KEY, leave LLM_BASE_URL empty for the default endpoint, and set LLM_FAST_MODEL and LLM_SMART_MODEL to model IDs from your provider.
4. Run `pnpm dev`, then open http://localhost:3000.
5. In Settings, under Model, you can also configure your own Anthropic or OpenAI compatible service; credentials stay in your browser.

### Prompt to paste into your agent

```text
Clone https://github.com/GeminiLight/SocialCoach, go to the app folder, install dependencies with pnpm, and configure .env.local for <provider> using <model ID>. Start the dev server and confirm that localhost:3000 loads and that one practice dialogue runs. Report any missing environment variables and never print the API key.
```

### Caveat

Conversation context is sent to the model provider you configure, so use your own credentials and check that provider's terms before entering personal situations.

Risk, credentials: Your model API key goes into .env.local or the browser settings, so keep it out of commits, logs, and screenshots.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/geminilight-socialcoach/run-socialcoach-locally-with-own-model-key/

## How to self-host a team instance with Docker Compose with Social Coach

Repo: [GeminiLight/SocialCoach](https://github.com/GeminiLight/SocialCoach) | Difficulty: Advanced | Time: 2 hours | Basis: From the README

Deploy a single-instance Social Coach behind Caddy with automatic HTTPS on your own server.

**Who it is for:** An ops-minded team lead or coach who wants a private Social Coach instance on a domain they control for a small group.

**The problem:** A shared group needs a hosted practice tool on its own domain rather than the public demo.

**The outcome:** The app is reachable over HTTPS at your domain and responds to a practice request.

### Steps

1. From the repository root, run `cd app` and then `cp .env.production.example .env.production`.
2. Edit .env.production with your model credentials and rate limit settings.
3. Replace example.com in the Caddyfile with your domain, point that domain's DNS to the server, and make sure ports 80 and 443 are open.
4. Run `docker compose up -d --build` from the app folder.
5. Open your domain in a browser and complete one practice dialogue to confirm the deployment.

### Prompt to paste into your agent

```text
In the SocialCoach repository, prepare a Docker Compose deployment in the app folder using <domain> and the provider settings in .env.production. Confirm that the Caddyfile uses the domain, that docker compose up -d --build succeeds, and that the site responds over HTTPS. Report the steps that need a human, such as DNS and secrets.
```

### Caveat

The README says the shared budget service only reserves anonymous tokens and does not guarantee prices, so set a hard spending limit with your model provider before opening the site to others.

Risk, credentials: The .env.production file holds shared model credentials, so restrict server access and set a provider spending cap as the README advises.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/geminilight-socialcoach/self-host-team-instance-with-docker-compose/