Preferred AI/Agent/Harnes local coding setups?

Frappe Framework App Development
,
1

What is everyone using for AI development?
I saw that @ankush mentioned Pi Agent. Is the landscape internally in Frappe diverse, or is there a coalescing around a specific toolset?

After hearing about PiAngent i gave that a try and REALLY like it so far. I’ve been using local Qwen3.8-27B models with pretty good success on local projects.

I’d be interested in seeing some discussion about what Models/Harnesses/MCPServers/Tools everyone is using.

Here’s my fully local setup I’ve so far settled on after LOTS of testing.

Model: Qwen3.8-27b
Inference Engine: Unsloth Studio
MCP Server Gateway: ToolHive
Editor: VSCode
Harness: Pi Agent

I Tried the following and had problems:

  • Continue.dev as harness - Continue.dev was bought out by Cursor(?)
  • KiloCode: works well - probably my second choice after Pi Agent
  • Most of the other options like OpenCode and proprietary harnesses seem to have too much bloat causing confusion for local-based models.
2

For Noviz AI Agent, we’ve kept the LLM layer flexible rather than tying it to a specific model.

  • LLMs: Qwen, Llama, DeepSeek, and other compatible models

  • Inference: Local/self-hosted or API-based

  • Agent/Harness: Custom Noviz agent runtime

  • Tools: MCP + ERPNext/Frappe APIs

  • Platform: ERPNext/Frappe

  • Deployment: Docker / self-hosted

The agent is designed to work across different LLMs while keeping the tool and ERP integration layer independent.

3

Thanks for sharing. I’ll check it out!
I see that it’s a paid app on Marketplace. Is the payment required if I’m using a local model?