DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to serve an open model to several coding agents at once with Paddock

Run one local model behind an OpenAI-compatible API that multiple agent sessions can share.

Authorization notice. Paddock is dual-use. Only use it on software, sites and systems you own or have explicit permission to test.

Who it is forA 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 problemRunning a separate model process for every agent session wastes GPU memory and makes long, repeated context slow and costly to serve.

The outcomeA local endpoint that accepts OpenAI chat completions and Anthropic messages calls, with paged KV cache, continuous batching and radix prefix caching across sessions.

What you need first

  • An NVIDIA GPU with the NVIDIA driver installed, or an Apple Silicon Mac running macOS 26 or later
  • Download access to the Paddock release page on GitHub (Windows, Linux, Linux arm64 on DGX Spark, or macOS)
  • Disk space for the model checkpoints you choose to download
  • For cloud comparison: an account with an OpenRouter or other OpenAI-compatible endpoint, if you want to compare against a cloud model

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 for your agent

Using the Paddock repo at truespar/paddock, configure my coding agent to use the locally served model at <local-server-address> 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.

Share: X LinkedIn Hacker News Email Copy link

Disclaimer

Deduped lists open-source repositories that belong to other people. We do not own, maintain, audit, or vouch for any of them, and a repo appearing here is not an endorsement or a safety check. The use cases, steps, and prompts are written by AI and may be incomplete, outdated, or just wrong. Where a use case is marked "Seen in the wild", we link to public posts where someone described something similar, but we have not verified those posts and the use case itself is still our AI's own write-up. Use all of this at your own risk. Only run code, commands, or agents on systems and accounts you are authorized to use, and read the code and every command before you run it, especially anything that touches credentials, files, or the network. Nothing here is professional, security, legal, or financial advice. All trademarks and project names belong to their respective owners, and their use here does not imply affiliation. Content is provided as is, without warranties of any kind. Found a mistake, a safety problem, or something you want removed? Email [email protected] and we will look at it promptly. Full disclaimer page