AUTOMATION

Build a Local Coding Agent with Distributed Tool Calling

Run a large local model that delegates tool calls to smaller models on edge devices

Updated: 5/20/2026
Difficulty
hard
Time
4-6 hours
Use Case
Execute local commands and code changes across multiple machines using a large model coordinator and smaller specialized models on edge devices
Popularity
0 views

About this automation

Deploy a local coding agent that uses a large model to plan and delegate tool calls to smaller models (functiongemma, Qwen) running on laptops or edge devices. Uses OATs Protocol's 141K+ tool index to standardize tool calling across heterogeneous hardware.

How to implement

1

Set up vLLM deployments for Qwen 27B/35B with tool-calling support

2

Deploy functiongemma on edge devices (old GPUs, laptops)

3

Create or customize OATs Tool-Calling Prompt Index JSON for your approved tools

4

Configure coder agent to load the prompt index on startup

5

Set up Mattermost instances for audit log posting and human review

6

Implement tool call monitoring and archiving to parquet files

7

Test with small command sets before running autonomous workflows

8

Archive execution logs for training and compliance