# AI use cases for Paddock (truespar/paddock) Repo: [truespar/paddock](https://github.com/truespar/paddock) | Use cases: 5 | Page: https://dedupedai.com/use-cases/truespar-paddock/ Paddock is a native Rust inference server that runs open AI models on NVIDIA GPUs and Apple Silicon, with OpenAI and Anthropic compatible APIs and a built-in web Studio. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Serve an open model to several coding agents at once: https://dedupedai.com/use-cases/truespar-paddock/serve-local-model-to-coding-agents/ - Compare a local model with a cloud model in Studio: https://dedupedai.com/use-cases/truespar-paddock/compare-local-and-cloud-models-side-by-side/ - Turn scanned pages and PDFs into text with OCR: https://dedupedai.com/use-cases/truespar-paddock/extract-text-from-scanned-pages/ - Fill a missing table column from labelled rows: https://dedupedai.com/use-cases/truespar-paddock/fill-missing-table-column-without-training/ - Transcribe a recording and show who spoke when: https://dedupedai.com/use-cases/truespar-paddock/transcribe-meeting-with-speaker-timeline/ ## How to serve an open model to several coding agents at once with Paddock Repo: [truespar/paddock](https://github.com/truespar/paddock) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README Run one local model behind an OpenAI-compatible API that multiple agent sessions can share. **Who it is for:** A developer or small team with one GPU box that wants several coding-agent sessions to use the same local model, with long contexts and shared prompt prefixes. **The problem:** Running a separate model process for every agent session wastes GPU memory and makes long, repeated context slow and costly to serve. **The outcome:** A local endpoint that accepts OpenAI chat completions and Anthropic messages calls, with paged KV cache, continuous batching and radix prefix caching across sessions. ### Steps 1. Install the NVIDIA driver on the machine, since that is the only requirement on NVIDIA systems. 2. Download the Paddock release for your platform from the GitHub releases page. 3. Run `paddock` to start the manager and open the browser-based Studio. 4. In the Studio, download a model from the Models tested list, such as a Qwen or Gemma 4 model, and start serving it. 5. Point each agent's OpenAI or Anthropic compatible base URL at the local server and run the sessions. ### Prompt to paste into your agent ```text Using the Paddock repo at truespar/paddock, configure my coding agent to use the locally served model at through its OpenAI-compatible chat completions API. Return the base URL and model name the agent should use. Verify by sending one short chat completion request and confirming a non-empty reply. ``` ### Caveat The README says the design target is several agent sessions on one box, but the aim of the project is single-GPU models, since tensor parallelism is not yet supported. 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/truespar-paddock/serve-local-model-to-coding-agents/ ## How to compare a local model with a cloud model in Studio with Paddock Repo: [truespar/paddock](https://github.com/truespar/paddock) | Difficulty: Beginner | Time: 45 minutes | Basis: From the README Send the same prompt to a model on your box and a model behind OpenRouter, then view the answers together. **Who it is for:** An AI engineer or product owner choosing between a self-hosted open model and a hosted model for a task. **The problem:** Judging whether a local model is good enough means running the same prompt through two different tools and comparing the output by hand. **The outcome:** A side-by-side view in the Studio showing the local and cloud answers to one prompt, so you can decide which model to use. ### Steps 1. Run `paddock` and open the Studio in your browser. 2. Download and start a model from the Models tested list on your own machine. 3. Add an OpenRouter or other OpenAI-compatible endpoint in the Studio as the cloud side of the comparison. 4. Enter one prompt and send it to both models in the side-by-side view. 5. Read both answers and note which one fits your task better. ### Prompt to paste into your agent ```text Using the Paddock repo at truespar/paddock, set up a side-by-side comparison in the Studio between a local model at and a cloud model at . Output the prompt and both replies. Verify that both answers appear in the same view for the same prompt. ``` ### Caveat The Studio is a web UI in the CLI package, and the native macOS app is pre-release and less complete than the web Studio. 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/truespar-paddock/compare-local-and-cloud-models-side-by-side/ ## How to turn scanned pages and PDFs into text with OCR with Paddock Repo: [truespar/paddock](https://github.com/truespar/paddock) | Difficulty: Beginner | Time: 45 minutes | Basis: From the README Read a page as plain text or as labelled blocks with their positions, using a local OCR model. **Who it is for:** An operations or records clerk with a folder of scanned PDFs or page images that need to become searchable text. **The problem:** Retyping scanned pages by hand is slow and error-prone, and cloud OCR may not be allowed for some documents. **The outcome:** Text output for each page, or labelled blocks with their boxes, produced on your own hardware. ### Steps 1. Install Paddock and run `paddock` to open the Studio. 2. Download LightOnOCR-3 (0.8B or 4B) or PaddleOCR-VL 1.6 from the Documents and OCR models in the Studio. 3. Open the PDF in the Studio, since PDF rendering to images is built in through pdfium. 4. Run the OCR model on the pages and choose plain text or labelled blocks as the output. 5. Save the output and check it against the original pages for errors. ### Prompt to paste into your agent ```text Using the Paddock repo at truespar/paddock, run the LightOnOCR-3 model on the PDF at and return the text of the first page. Verify by comparing the output with the page visually for any obvious missed or wrong words. ``` ### Caveat The README lists PDF rendering as an optional convenience, so the quality of OCR output still needs checking by a person on important documents. 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/truespar-paddock/extract-text-from-scanned-pages/ ## How to fill a missing table column from labelled rows with Paddock Repo: [truespar/paddock](https://github.com/truespar/paddock) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README Paste a table with labelled rows and let Kumo Tabular predict the missing column for new rows. **Who it is for:** A data analyst with a small spreadsheet of labelled records where one column is blank for new rows. **The problem:** Building a model to predict a missing column usually means training code, which is too much work for a one-off table. **The outcome:** Predicted values for the missing column of the new rows, produced in the Studio Tables page with no training step. ### Steps 1. Run `paddock` and open the Studio. 2. Download a Kumo Tabular model (small, medium or large, for classification or regression) from the Studio. 3. Open the Tables page and paste or open your table, with the labelled rows and the new rows. 4. Run the prediction to fill in the missing column for the new rows. 5. Review the predictions and check a few against known values before using them. ### Prompt to paste into your agent ```text Using the Paddock repo at truespar/paddock, send the labelled rows in to the /v1/tabular/predictions endpoint with Kumo Tabular and return the predicted values for the unlabelled rows. Verify by checking that the number of predictions equals the number of unlabelled rows. ``` ### Caveat The README describes Kumo Tabular as needing no training, but it does not state accuracy, so predictions should be checked before they drive decisions. 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/truespar-paddock/fill-missing-table-column-without-training/ ## How to transcribe a recording and show who spoke when with Paddock Repo: [truespar/paddock](https://github.com/truespar/paddock) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README Get a transcript and a speaker timeline for a recording of up to 10 minutes on your own machine. **Who it is for:** A researcher, journalist or team lead who needs a written record of a short interview or meeting with clear speaker turns. **The problem:** Sorting out who said what in a transcript is tedious, and uploading recordings to a hosted service may not suit the content. **The outcome:** A transcript with speaker turns shown on a timeline, including overlapping speech, produced locally. ### Steps 1. Run `paddock` and open the Studio. 2. Download Qwen3-ASR 1.7B for transcription and Nemotron 3 Diarization for speaker turns. 3. Open the recording in the Studio, keeping it under 10 minutes for diarization. 4. Run transcription and diarization on the recording. 5. Read the speaker timeline and correct any turns that look wrong. ### Prompt to paste into your agent ```text Using the Paddock repo at truespar/paddock, transcribe the audio file and run speaker diarization on it. Return the transcript with speaker labels. Verify that the audio is under 10 minutes long before running diarization. ``` ### Caveat The README says diarization works on recordings up to 10 minutes, so longer meetings will need to be split first. 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/truespar-paddock/transcribe-meeting-with-speaker-timeline/