# AI use cases for Talk with Me (scorbo2/TalkWithMe) Repo: [scorbo2/TalkWithMe](https://github.com/scorbo2/TalkWithMe) | Use cases: 5 | Page: https://dedupedai.com/use-cases/scorbo2-talkwithme/ TalkWithMe is a local single-user web chat app that connects to a llama.cpp server and supports multi-persona group chats with optional text-to-speech and speech-to-text. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Run a private group chat with AI personas: https://dedupedai.com/use-cases/scorbo2-talkwithme/private-group-chat-with-personas/ - Hear and speak to personas with TTS and STT: https://dedupedai.com/use-cases/scorbo2-talkwithme/voice-persona-chat-with-tts-and-stt/ - Point Talk with Me at a remote LLM with an API key: https://dedupedai.com/use-cases/scorbo2-talkwithme/chat-through-a-remote-llm-with-api-key/ - Let personas call tools from MCP servers: https://dedupedai.com/use-cases/scorbo2-talkwithme/give-personas-mcp-tools/ - Build a persona library by cloning and renaming: https://dedupedai.com/use-cases/scorbo2-talkwithme/clone-and-rename-personas-library/ ## How to run a private group chat with AI personas with Talk with Me Repo: [scorbo2/TalkWithMe](https://github.com/scorbo2/TalkWithMe) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README Run a fully local multi-persona chat on your own llama.cpp server with no internet or login. **Who it is for:** A developer or hobbyist with a local LLM who wants several AI characters to talk in one room without sending chats to a cloud service. **The problem:** Single-model chat tools offer one voice at a time and often depend on hosted services. **The outcome:** A browser chat at localhost where one or more personas answer in a shared room, with messages saved per room. ### Steps 1. Start your llama.cpp server with its OpenAI-compatible API enabled, using the --api flag. 2. In the project folder, run `pip install -r requirements.txt`. 3. Run `uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload`. 4. Open http://localhost:8000 in your browser, then open Settings and confirm the LLM base URL (the README example is http://localhost:8080). 5. Create a chat room, assign two or more personas to it, and choose a routing mode such as LLM-decided, random, or manual. ### Prompt to paste into your agent ```text Use the TalkWithMe repo to start a local chat at http://localhost:8000 against a llama.cpp server at . Create one room with the personas Alex and Luna, send the message , and report the persona replies. Verify that the room's messages are still listed after a page reload. ``` ### Caveat The app needs a separately running llama.cpp server, which the README does not set up for you. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/scorbo2-talkwithme/private-group-chat-with-personas/ ## How to hear and speak to personas with TTS and STT with Talk with Me Repo: [scorbo2/TalkWithMe](https://github.com/scorbo2/TalkWithMe) | Difficulty: Advanced | Time: 120 minutes | Basis: From the README Add spoken replies from a TTS server and voice prompts through a Whisper-style STT server. **Who it is for:** A user who wants to talk to AI personas by voice and hear them reply, usually on a home workstation with a GPU. **The problem:** Text-only chat is slow for conversation practice, accessibility, or hands-free use. **The outcome:** Persona replies play as audio in the room, and you can click the microphone icon to speak your prompt. ### Steps 1. Start a tts-serve instance and note its base URL, then set tts.base_url in settings.yaml (the README example is http://localhost:5500). 2. Start an OpenAI-compatible STT server with a /v1/audio/transcriptions endpoint, such as whisper-fastapi, and set stt.base_url to its root (the README example is http://localhost:8181). 3. Set tts.enabled and stt.enabled to true in settings.yaml. 4. Run `uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload` and open http://localhost:8000. 5. Click the microphone icon, speak a prompt, and confirm that a persona replies with audio. ### Prompt to paste into your agent ```text Use the TalkWithMe repo to configure TTS at and STT at in settings.yaml, then start the app. Send the spoken test phrase and report the transcribed text and whether a persona audio reply was produced. Verify that the transcription matches the phrase. ``` ### Caveat TTS and STT are separate servers that you must install and run yourself; the README points to tts-serve and whisper-fastapi for them. Risk, voice cloning: Persona folders can hold ref.wav reference audio for voice cloning, so only use voices you have permission to clone. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/scorbo2-talkwithme/voice-persona-chat-with-tts-and-stt/ ## How to point Talk with Me at a remote LLM with an API key with Talk with Me Repo: [scorbo2/TalkWithMe](https://github.com/scorbo2/TalkWithMe) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Connect the chat app to a hosted OpenAI-compatible LLM while keeping the key out of settings.yaml. **Who it is for:** A user whose local hardware is too small for a good model and who wants to use a hosted LLM from a provider they trust. **The problem:** Local models may be too slow or weak, and the app needs a credential that must not sit in a config file. **The outcome:** Chats go to the remote LLM, with the Authorization header sent on each request and the key never shown in the UI or logs. ### Steps 1. Run `cp llm_api_key.example llm_api_key` in the project root. 2. Open the llm_api_key file and replace your_key_here with your API key. 3. Run `chmod 600 llm_api_key` to keep the key readable only by you. 4. Set llm.base_url in settings.yaml to your provider's OpenAI-compatible URL, using https://. 5. Run `uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload` and send a test message. ### Prompt to paste into your agent ```text Use the TalkWithMe repo to connect to the remote LLM at using the key in llm_api_key. Send the test message and report the reply and the startup log line saying whether a key is configured. Verify that the key value itself does not appear in the log or the UI. ``` ### Caveat The README warns that only remote LLMs you trust should be connected, and that cleartext http:// URLs expose chats and the key. Risk, credentials: Your API key is sent in each LLM request, and remote usage may incur costs, so keep llm_api_key out of version control and watch your bill. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/scorbo2-talkwithme/chat-through-a-remote-llm-with-api-key/ ## How to let personas call tools from MCP servers with Talk with Me Repo: [scorbo2/TalkWithMe](https://github.com/scorbo2/TalkWithMe) | Difficulty: Advanced | Time: 60 minutes | Basis: From the README Enable tool calls so personas can fetch web pages or run queries through Streamable HTTP MCP servers. **Who it is for:** A developer who wants personas to act on live data, such as reading a web page or querying a database, through servers they run or trust. **The problem:** A plain chat model cannot read current web pages or query your own data sources. **The outcome:** Personas with tool calls enabled can use configured MCP tools during a reply, and the app can show those tool calls. ### Steps 1. Start one or more MCP servers that expose the Streamable HTTP transport. 2. Add each server to the mcp.servers list in settings.yaml, and keep max_tool_iterations at the default of 8 unless you need a change. 3. Restart the app, because the mcp section is only read on startup. 4. Open the Personas editor and set allow_tool_calls to true for the persona that should use tools. 5. Run `uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload`, then ask the persona to fetch a web page and check the tool call in the chat. ### Prompt to paste into your agent ```text Use the TalkWithMe repo to configure MCP server in settings.yaml, enable tool calls for the persona , and restart the app. Ask the persona to use the tool on and report the tool call shown in the chat. Verify that the tool result appears in the reply. ``` ### Caveat The README says the mcp section has no UI and must be edited in settings.yaml, with a restart after each change. Risk, remote script: Tool calls run on the MCP servers you list, so only add servers you trust and that you have checked. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/scorbo2-talkwithme/give-personas-mcp-tools/ ## How to build a persona library by cloning and renaming with Talk with Me Repo: [scorbo2/TalkWithMe](https://github.com/scorbo2/TalkWithMe) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README Clone a stock persona, rename it, and keep the on-disk folder matched to the new name. **Who it is for:** A user who wants a set of distinct AI characters for different rooms and would rather adapt existing personas than write each from scratch. **The problem:** Creating many personas by hand is slow, and copies often leave numbered folders behind. **The outcome:** A persona library in the Personas folder, where renamed clones usually have directories that match their new names. ### Steps 1. Run `pip install -r requirements.txt` and `uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload` to start the app. 2. Open http://localhost:8000 and select the Personas control in the top right. 3. Select Clone on the Alex persona, which adds a numeric suffix such as Alex_2 to the name. 4. Edit the clone's name, description, router_hints, and system prompt inline. 5. Check the Personas folder on disk to confirm the directory name matches the new display name, and check the name: line in prompt.md if it does not. ### Prompt to paste into your agent ```text Use the TalkWithMe repo to clone the persona Alex, rename the clone to , and update its description to . Report the new directory name under Personas/ and the contents of its prompt.md. Verify that the directory name matches the new name or that a name: line records it. ``` ### Caveat Directory renames are best-effort, so names with no usable characters or with collisions keep the old directory name. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/scorbo2-talkwithme/clone-and-rename-personas-library/