# LLMs for Research ## Inference on Demand This page orients KSU researchers to using self-service LLM API keys with BYOK-compatible apps while explaining how provisioning, security, and model access work through Parley. ## Quick overview * **Who**: KSU researchers who need LLM access for scholarship and prototyping. * **What**: A single, email-bound API key that authenticates you to Parley’s gateway to OpenAI and other models. * **Where**: Create or rotate your key at using your @kennesaw\.edu email. * **Why**: Secure access, usage tracking, built-in moderation guardrails, and a curated set of model families (text reasoning, embeddings, and image). * **How**: Use your Parley key inside any BYOK application; this page links to tutorials for common tools and shows the standard connection settings. # What is Parley? Many research tools support Bring Your Own Key (BYOK)—you provide an API key, the app routes requests through your institution’s gateway, and you keep control over usage. With Parley, your key: * Is provisioned to your KSU email and is one-active-key-at-a-time. * Authenticates through Parley’s secure gateway, not the public internet directly. * Enables per-user usage tracking and budgeting for research stewardship. * Applies LLM-based Moderation automatically as a guardrail before requests are forwarded. ## Get your Parley key 1. Go to . 2. Sign in with your KSU email and request a key. 3. Use the magic link in your email to generate the key in a web browser. 4. Copy your key and store it securely. 5. (Optional) If you already had a key, this process rotates it out. Only the newest key remains active. > **Note:** Keys are provided to support research **as resources permit**. Availability and rate limits may adjust over time. ## Why use Parley keys? **Security & Compliance** * All traffic runs through a secure access channel with auditability. * Moderation is applied by default to reduce risk and support responsible use. **Accountability & Budgeting** * Keys are tied to individual researchers for accurate usage tracking. * Visibility helps align compute spend with research priorities. **Model Choice** * Parley exposes a curated catalog including: * Reasoning language models for analysis, drafting, and tool-use. * Embedding models for search, clustering, RAG, and semantic analytics. * Image models for generation and editing where appropriate. * Tutorials indicate which model names to select in each app. ## Choosing Your Application To make use of your Parley key, you can choose an application (and walkthrough) from these options.

			Cover image
Cherry Studio	Supported Platforms: Windows, Mac, Linux	Useful Services: Chat, Images, Chat with Files	icon.png	Cherry Studio
Codename Goose	Supported Platforms: Windows, Mac, Linux	Useful Services: Chat, Task Automation, Chat with Files	goose-logo-black.png	Codename Goose
Apollo by Liquid	Supported Platforms: iOS, Android, Mac	Useful Services: Chat, Chat with Files	GXuYC0tu_400x400.jpg
Witsy	Supported Platforms: Windows, Mac, Linux	Useful Services: Chat, Chat with Files	witsy.png

## Responsible Use * Every request flows through the OpenAI Moderation API that flags or blocks content violating institutional policy. * If moderation intervenes, the app will return an explanatory message. Adjust your prompt or consult the guidance section of the tutorial. * Follow all applicable data handling policies; do not paste regulated or sensitive data (such as PII) unless your workflow explicitly allows it. ## Frequently Asked Questions **Can I have multiple keys?**\ No. **Only one key can be active at a time**, bound to your KSU email. Rotating a key deactivates the old one. **What if my key is exposed or leaked in code?**\ Rotate it immediately at and update your apps. Avoid committing keys to repos; use environment variables or secret managers. **Are there usage limits?**\ Yes. Parley manages quotas to align with available resources and research priorities. Tutorials note any current rate limits. **Which models are available right now?**\ We list current reasoning, embedding, and image models in the model catalog on each tutorial. Model names may evolve; always check the tutorial’s “Latest Models” box. **Can I use this in pipelines or notebooks?**\ Yes. The same key works in notebooks, RAG services, and CLI tools—any OpenAI-compatible client that supports custom base URLs. ## Support If you encounter configuration errors, moderation false-positives, or model selection questions, contact Dylan Goldblatt (). He provides support for key provisioning, quotas, and policy questions. # Cherry Studio It takes 5 minutes to integrate your Parley API key with Cherry There are numerous ways to make use of your Parley credentials. This guide will show you one of the easiest ways to get started with multiple types of functionality, using a laptop or desktop in the operating system of your choosing. {% embed url="" %} {% stepper %} {% step %} ## Install Cherry Studio The highlighted options are the most likely matches for your work computer. Note: If you have an older Mac with an Intel processor, select the x64 version. Linux packages are available, too. These versions change as the software is updated, so you may need to locate these extensions on the latest release.\ \ **Official Cherry AI Website:**\ \ \ **Page with the latest releases:**\
* Example Windows Installer Packages (x64-setup.exe) Look for something like * Example macOS Installer Packages (arm64.dmg) Look for something like {% endstep %} {% step %} ## Open the Settings Menu Use the settings gear icon to enter the configuration screen.

{% endstep %} {% step %} ## Open the Model Provider Menu In the top of the left rail, select the Model Provider tab. {% endstep %} {% step %} ## Remove CherryIN Provider By default, this app includes a provider that is not part of the Parley service. Please disable CherryIN by turning off the green switch in the top right of the settings screen.

{% endstep %} {% step %} ## Add a Model Provider In the bottom of the second column, click the + Add button.

{% endstep %} {% step %} ### Insert Parley API Key Now it is time to paste the key you collected from into the API Key form.

{% endstep %} {% step %} ### Insert Parley API Host Enter the host URL in the API Host form **precisely** as follows. ``` https://keys.theparley.org/v1 ``` {% endstep %} {% step %} ### Add Models From the List Under the *Models* header, select *Add*. This will load the models active on the key. Find models such as *gpt-5.2* and *gpt-image-1* in the list.

{% endstep %} {% step %} ### Confirm GPT-5.2 is Functioning Select the *Check* button in the API Key area, select *gpt-5.2* and select *OK.* You will see the green check and "Connection Successful" message if all works well.

{% endstep %} {% step %} ### Set GPT-5.2 as the Default Open the Default Model menu from the left rail and ensure that gpt-5 is the default for all.

{% endstep %} {% step %} ## Chat with Parley Select the *Home* button and start conversing with the model *gpt-5.2.*

{% endstep %} {% step %} ## Create Images with Parley Select a diffusion model (e.g., *gpt-image-1*) in a new chat. Prompts create images.

{% endstep %} {% endstepper %} # Codename Goose This guide shows how to integrate your Parley API key into Codename Goose using only the GUI, so you can immediately access chat and image generation features powered by Parley. {% stepper %} {% step %} ## Install Codename Goose The linked options on Mac and PC are the most likely matches for your work computer. Linux packages are available, too. {% embed url="" %} {% endstep %} {% step %} ## Configure LLM Provider * [ ] Open Goose Desktop. * [ ] Navigate to Settings or Preferences (usually via a gear icon or menu bar). * [ ] Find a section labeled Providers or LLM Provider Configuration. * [ ] Select OpenAI as the LLM Provider and configure it. * [ ] Request a Parley API key through the self-service portal, . * [ ] In the provider setup form, enter these and hit submit. * **Name**: `Parley` * **API Key**: your unique API key * **API Host / Endpoint**:\ `https://keys.theparley.org/v1` * **OpenAI Base Path**\ `v1/chat/completions`

{% endstep %} {% step %} ## Enable Parley Models * [ ] Still in the provider configuration section, locate a subsection for Models or Available Models. * [ ] From the list (which Goose pulls via your key), identify and add the Parley models you intend to use. Example: `gpt-5` (text / chat). * [ ] Once added, a notification should appear to confirm that it has been configured.

{% endstep %} {% step %} ## Chat in Goose * [ ] Return to Goose’s main window (e.g. Home or Chat tab). * [ ] Start a new chat session. Goose will route requests via Parley using `gpt-5` by default. * [ ] Enter your prompt and receive responses as you would in any chat interface. {% endstep %} {% step %} ## Optional: Add Extensions * [ ] Select the *Extensions* menu in the left rail. * [ ] Activate the features that you find useful for your task. These are also known as MCPs.

{% endstep %} {% endstepper %} # Mobile Apps Catalog of apps that work on mobile devices with your Parley keys. * Apollo (iOS/Android) * Download: * Name: Parley * Base URL: * Chat Path: /v1/chat/completions * Models Path: /v1/models * Auth: {your API key} * Enchanted (MacOS, iOS, Apple Watch, Vision Pro) * Download: * Ollama server uri: * Bearer Token: {your API key} * Geeps (MacOS, iOS, iPadOS) * Download: * Configuration: OpenAI * Use custom host: {activate} * API Host e.g. api.openai.com: keys.theparley.org * API Key: {your API key} * Note: PDF attachments are a paid upgrade # Knowledge Base Tutorial Below, we will present detailed instructions for using Cherry Studio step-by-step. {% hint style="warning" %} This document was prepared by the Cherry AI team in Chinese and copied here for convenience {% endhint %} ### Add Embedding Model 1. Find the model in the model management service, you can quickly filter by clicking "Embedding Model"; 2. Find the required model and add it to My Models. 3. For Parley API users, this will be text-embedding-3-large.\ \ ![](https://544318696-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fsl8DR5pyUhDgjzEnE7ZK%2Fuploads%2FQox5kLNMGrc76g3IhVZR%2FScreenshot%202025-11-21%20at%201.19.33%E2%80%AFPM.png?alt=media\&token=dfeb8aba-de2c-405c-ad65-7ea281e6036d) ### Create Knowledge Base 1. Knowledge Base Entry: On the left toolbar of Cherry Studio, click the knowledge base icon to enter the management page; 2. Add Knowledge Base: Click "Add" to start creating a knowledge base; 3. Name: Enter the name of the knowledge base and add an embedding model. In this case, we recommend text-embedding-3-large. ### Add Files and Vectorize 1. Add Files: Click the "Add files" button to open the file selection; 2. Select Files: Choose supported file formats, such as pdf, docx, pptx, xlsx, txt, md, mdx, etc., and open them; 3. Vectorization: The system will automatically perform vectorization. When it shows "Completed" (green ✓), it means vectorization is finished. ### Add Data from Multiple Sources Cherry Studio supports multiple ways to add data: 1. Folder Directory: You can add an entire folder directory, and supported format files within that directory will be automatically vectorized; 2. Website Link: Supports URL links, such as [https://llm.theparley.org](< https://llm.theparley.org >). 3. Plain Text Note: Supports entering custom content in plain text. {% hint style="info" %} Tips: 1. Illustrations in documents imported into the knowledge base currently do not support conversion to vectors and need to be manually converted to text; 2. Using a website as a knowledge base source may not always succeed. Some websites have strict anti-crawling mechanisms (or require login, authorization, etc.), so this method may not always obtain accurate content. It is recommended to test with a search after creation. {% endhint %} ### Import Past Chats from ChatGPT {% embed url="" %} ### Search Knowledge Base Once files and other data are vectorized, you can query them: 1. Click the "Search Knowledge Base" button at the bottom of the page; 2. Enter the query content; 3. The search results will be displayed; 4. And the matching score for each result will be shown. ### Cite Knowledge Base in Conversation to Generate Replies 1. Create a new topic. In the conversation toolbar, click "Knowledge Base," which will expand the list of created knowledge bases. Select the knowledge base you want to cite; 2. Enter and send your question, and the model will return an answer generated from the retrieval results; 3. At the same time, the cited data source will be attached below the answer, allowing quick access to the source file. # Lunch & Learn Series ETA: January 2026, Office of Research * Part 1: Intro to Generative AI * What is ChatGPT? * What is Parley? * What is Generative AI good for? * Demystify capabilities/limits; prompt basics; responsible use * Part 2: Using Parley at Work for Operations * Part 3: Using Parley for Research Tasks * Part 4: Applying Parley to Your Role

	Turntable Team	BOOMBOX CREW	LIMEWIRE LEGENDS	SPOTIFY SQUAD
Week 1
Week 2
Week 3
Week 4

# Week 1 Show immediate value. ## AI Foundations & First Wins ### MONDAY — Turntable Team Part 1: AI Foundations (0-30 min) * What is generative AI and how does it work (simple explanation) * What AI is good/bad at (strengths and critical limitations) * Hallucinations and accuracy concerns * Privacy, data security, and institutional policies * When to use AI vs. when not to * Basic prompt writing principles ### Part 2: Hands-On Practice (30-60 min) * **Activity**: Take a messy, multi-party email chain and transform it into: * A one-page executive brief * Key risks and opportunities * Decision-ready talking points * Participants work individually or in pairs using AI tools * Share results and discuss: What worked? What didn't? How do you verify accuracy? * Instructor provides real-time coaching and refinement tips **Homework**: Try one AI task from your real work this week. Document what happened. ### TUESDAY — BOOMBOX CREW ### **Part 1: AI Foundations (0-30 min)** * What is generative AI, and how does it work * Strengths and limitations for professional work * Recognizing when AI output is unreliable * Data privacy and institutional guidelines * Appropriate vs. inappropriate use cases * Introduction to effective prompting ### **Part 2: Hands-On Practice (30-60 min)** * **Activity**: Summarize a complex policy or guideline into: * A one-page clear summary * A ready-to-send email explanation * Participants use AI tools to complete the task * Group shares outputs and evaluates: Is this accurate? How do we check? * Instructor guides prompt refinement and troubleshooting **Homework**: Apply AI to one recurring task. Note successes and failures. ### WEDNESDAY — LIMEWIRE LEGENDS ### **Part 1: AI Foundations (0-30 min)** * How generative AI works (accessible overview) * What it excels at and where it fails * Understanding and catching AI mistakes * Privacy considerations and workplace policies * Developing judgment about when to use AI * Basics of writing good prompts ### **Part 2: Hands-On Practice (30-60 min)** * **Activity**: Transform a sensitive or overwhelming email thread into: * A diplomatic, well-structured response * A clear next-steps plan * Participants actively use AI to complete transformations * Share and compare results: How would you verify tone and accuracy? * Real-time feedback and prompt improvement **Homework**: Test AI on one real work challenge. Bring results to share. ### THURSDAY — SPOTIFY SQUAD ### **Part 1: AI Foundations (0-30 min)** * What generative AI is and how it functions * Key strengths and important limitations * Spotting unreliable or incorrect outputs * Data security and organizational policies * Building judgment: when AI helps vs. when it doesn't * Getting started with prompts ### **Part 2: Hands-On Practice (30-60 min)** * **Activity**: Turn a messy email thread or assignment into: * An organized action plan * A simplified, confident summary * Participants work with AI tools to create outputs * Group discussion: How do you know if the output is good? * Coaching on improving results **Homework**: Use AI for one task this week. Document the experience. ### FRIDAY - OFFICE HOURS Answer any questions from the different cohorts. Only one hour? Maybe two?
### KEY TAKEAWAYS?

Turntable Team	BOOMBOX CREW	LIMEWIRE LEGENDS	SPOTIFY SQUAD

# Week 2 Create repeatable templates that solve their recurring workflow problems. Each group leaves with 2–3 customized prompt “recipes.” ## Prompt Recipes: Build Your Own Mini-Assistant *** ### MONDAY — Turntable TEAM **Part 1: Prompt Engineering + Homework Debrief (0-30 min)** * Share homework experiences: what worked, what failed, why * Anatomy of effective prompts (role, context, task, format, constraints) * Iterative refinement techniques * Getting better outputs through specificity * Common prompt mistakes and fixes **Part 2: Build Your Recipe Library (30-60 min)** * **Activity**: Create 2-3 custom prompt recipes: * Executive brief generator * Talking points creator * "Explain this issue for leadership" template * Strategic-risk summary * Participants actively build and test each recipe using AI * Refine prompts based on results * Save successful recipes in a shared document * Share best recipes with the group **Outcome**: Practical templates ready for immediate leadership communication needs. ### TUESDAY — BOOMBOX CREW **Part 1: Prompt Engineering + Homework Debrief (0-30 min)** * Homework review: lessons learned from real applications * How to structure prompts for consistent results * Iteration strategies for accuracy * Using examples to guide AI output * Troubleshooting poor responses **Part 2: Build Your Recipe Library (30-60 min)** * **Activity**: Create 2-3 custom prompt recipes: * Policy → plain English translator * Structured memo generator * Decision-tree extractor * Minutes → decisions + deadlines * Participants use AI to develop and test templates * Iterate to improve quality and consistency * Build shared recipe collection * Present favorite recipes to group **Outcome**: Reliable templates for documentation and process clarity. ### WEDNESDAY — LIMEWIRE LEGENDS **Part 1: Prompt Engineering + Homework Debrief (0-30 min)** * Share real-world attempts: successes and failures * Building prompts that deliver consistent quality * How to iterate when results aren't quite right * Tone and style guidance in prompts * Fixing common prompt problems **Part 2: Build Your Recipe Library (30-60 min)** * **Activity**: Create 2-3 custom prompt recipes: * Project update generator * Committee summary template * Stakeholder briefing drafts * Rewrite for tone ("clear, kind, professional") * Participants actively create and test recipes with AI * Refine based on actual outputs * Compile group recipe library * Share best practices and examples **Outcome**: Templates that reduce repetitive communication work. ### THURSDAY — SPOTIFY SQUAD **Part 1: Prompt Engineering + Homework Debrief (0-30 min)** * Homework review: what you tried and what you learned * What makes a prompt effective * How to improve outputs through refinement * Giving AI the right context and examples * Diagnosing why outputs miss the mark **Part 2: Build Your Recipe Library (30-60 min)** * **Activity**: Create 2-3 custom prompt recipes: * Meeting recap generator * Instruction clarifier * Professional email drafts * Turn vague tasks into step-by-step plans * Participants build and test templates using AI * Practice and troubleshoot together * Build shared template collection * Demonstrate successful recipes **Outcome**: Organizational tools for clarity and confidence. ### FRIDAY - OFFICE HOURS Answer any questions from the different cohorts. Only one hour? Maybe two? ### KEY TAKEAWAYS?

Turntable Team	BOOMBOX CREW	LIMEWIRE LEGENDS	SPOTIFY SQUAD

# Week 3 Learn to use Parley (internal researcher database) to find experts quickly, then use AI to synthesize and communicate findings. ### Parley for Research Discovery + AI Synthesis #### MONDAY — TURNTABLE TEAM **Part 1: Parley Navigation Tutorial (0-30 min)** * Introduction to [Parley:](/what-is-parley) what it is and how to search it * Finding researchers by topic, expertise, or center * Understanding research profiles and comparing capabilities * Common search strategies * Demo: instructor walks through a search example **Part 2: Strategic Research Matching (30-60 min)** * **Activity**: Given an external partnership request: * Participants use Parley to identify 3-5 aligned researchers * Compare their strengths, overlaps, and gaps * Use AI to create a one-page strategic recommendation with talking points * Work individually or in small groups * Share findings and AI-generated recommendations * Practice checking AI accuracy against Parley data * Instructor provides feedback on search strategies and synthesis quality **Outcome**: Parley + AI becomes a decision-making tool for partnerships. #### TUESDAY — BOOMBOX CREW **Part 1: Parley Navigation Tutorial (0-30 min)** * How to use Parley effectively * Search techniques for finding the right researchers * Reading and comparing research profiles * Tips for accurate searches * Live demonstration of a typical search workflow **Part 2: Expertise Mapping & Documentation (30-60 min)** * **Activity**: Given a research domain: * Participants use Parley to identify 4-6 researchers working in similar areas * Build a comparison table (expertise, center, focus areas) * Use AI to rewrite complex bios into clear internal summaries * Work on real scenarios individually or in pairs * Share comparison tables and summaries * Verify AI summaries against source profiles as a group * Discuss accuracy checks and quality standards **Outcome**: Parley + AI clarifies research roles and responsibilities. #### WEDNESDAY — LIMEWIRE LEGENDS **Part 1: Parley Navigation Tutorial (0-30 min)** * Parley overview and search fundamentals * Matching inquiries to researcher expertise * Understanding profile information * Efficient search workflows * Walkthrough of common use cases **Part 2: Communication & Routing (30-60 min)** * **Activity**: Given a research inquiry from a partner or funder: * Participants use Parley to match inquiry to 2-3 appropriate faculty * Draft a professional outreach email using AI * Create a simple summary of each researcher's focus for stakeholders * Complete the full workflow hands-on * Share emails and summaries with group * Group review: tone, accuracy, and clarity checks * Practice verifying details before sending **Outcome**: Parley + AI supports fast, accurate communication and routing. #### THURSDAY — SPOTIFY SQUAD **Part 1: Parley Navigation Tutorial (0-30 min)** * Introduction to Parley and why it's useful * Basic search strategies * How to interpret researcher profiles * Finding who handles specific research areas * Step-by-step demo of typical searches **Part 2: Orientation & Resource Building (30-60 min)** * **Activity**: Create onboarding resources: * Participants use Parley to answer "Who handles X research area?" * Turn 3-4 researcher profiles into simple FAQs using AI * Create a "Who to Contact" quick reference guide * Work individually or in small teams * Share and test resources with each other * Verify accuracy and usefulness * Refine based on peer feedback **Outcome**: Parley + AI helps newcomers understand the organization quickly. ### FRIDAY - OFFICE HOURS Answer any questions from the different cohorts. Only one hour? Maybe two? ### KEY TAKEAWAYS?

Turntable Team	BOOMBOX CREW	LIMEWIRE LEGENDS	SPOTIFY SQUAD

# Week 4 Create a simple AI-powered tool, present it to your cohort, and learn from everyone's real-world experiences. \*Note: We'll use PartyRock\* ### Build Your AI Tool + Share & Learn #### MONDAY — TURNTABLE TEAM **Part 1: Tool-Building Tutorial (20-25 min)** * Quick intro to the platform * How to set up a simple AI assistant with custom instructions * Demo: building a basic tool from scratch step-by-step * Overview of tool options and examples **Part 2: Build & Present (65-70 min)** *Build Phase (45 min):* * **Tool options**: * Leadership briefing generator * Strategic update assistant * Multi-document synthesizer * \[Or propose your own based on your needs] * Participants actively build their tool using the platform * Work individually or in pairs * Instructors circulate to help troubleshoot * Test tools with real examples *Share & Learn (20 min):* * 2-minute demo from each person/pair * Focus: "What problem does this solve and how would you use it?" * Group feedback and suggestions **Reflection**: What will you continue using? What did you learn about AI's capabilities and limits? #### TUESDAY — BOOMBOX CREW **Part 1: Tool-Building Tutorial (20-25 min)** * Platform walkthrough * Setting up custom AI assistants * Example: building a simple tool together live * Review of possible tool types **Part 2: Build & Present (65-70 min)** *Build Phase (45 min):* * **Tool options**: * Policy simplifier * Committee briefing generator * Instruction clarity assistant * \[Or design your own] * Participants create their tool on the platform * Individual or partner work * Support available throughout * Practice using tool with actual content *Share & Learn (20 min):* * Quick presentations from each participant * "What does this tool do and why is it useful?" * Peer feedback and ideas **Reflection**: Which tools will you actually use? What surprised you about this process? *** #### WEDNESDAY — LIMEWIRE LEGENDS **Part 1: Tool-Building Tutorial (20-25 min)** * Platform introduction * Creating customized AI assistants * Live demo of tool creation process * Examples of effective tools **Part 2: Build & Present (65-70 min)** *Build Phase (45 min):* * **Tool options**: * Project update assistant * Automated follow-up generator * Event communication helper * \[Or create based on your workflow] * Participants build and test their tool * Work individually or with partners * Help available for technical issues * Run through real scenarios *Share & Learn (20 min):* * Everyone presents their tool briefly * "Here's the problem it solves and how I'd use it" * Group discussion and refinement ideas **Reflection**: What tools will stick? What did you learn about AI integration in daily work? *** #### THURSDAY — SPOTIFY SQUAD **Part 1: Tool-Building Tutorial (20-25 min)** * Getting started with the platform * How to build a custom AI assistant * Walkthrough of a sample tool creation * Ideas for useful tools **Part 2: Build & Present (65-70 min)** *Build Phase (45 min):* * **Tool options**: * Onboarding assistant * "Explain this policy simply" tool * Recurring email helper * \[Or build something you need] * Participants create their tool actively * Individual or partner work * Guidance available throughout * Test with real examples *Share & Learn (20 min):* * Short presentations from all participants * "This is what it does and why it helps" * Feedback and improvement suggestions **Reflection**: What will you keep using? What have you learned about working with AI? Pitch Angle:\ “How this tool helps new staff get organized quickly.” ### FRIDAY - OFFICE HOURS Answer any questions from the different cohorts. Only one hour? Maybe two? ### KEY TAKEAWAYS?

Turntable Team	BOOMBOX CREW	LIMEWIRE LEGENDS	SPOTIFY SQUAD