- Guides
- Building Agentic Coding Workflows
Guides
Building Agentic Coding Workflows
Use North Mini Code in tool-using agent loops that edit code, inspect repos, and operate from the terminal.
North Mini Code was trained specifically for agentic software engineering. It performs best when the surrounding runtime handles tools, tool results, and multi-step conversation state cleanly.
Describe tools with JSON schema
North Mini Code expects tools to be described as functions with explicit JSON-schema parameters.
Render the chat with the model's template
Use the tokenizer chat template so tool definitions and assistant turns are formatted the way the model expects.
Execute the tool outside the model
The model proposes a tool call. Your runtime executes the shell command, file edit, or API request.
Append both tool output and model reasoning back into history
The Hugging Face model card says North Mini Code supports interleaved thinking and works best when both the thinking content and tool results are carried into future steps.
Tool definition example
Conversation loop
tools = [
{
"type": "function",
"function": {
"name": "bash",
"description": "Execute a bash command in the terminal.",
"parameters": {
"type": "object",
"properties": {
"command": {
"type": "string",
"description": "The bash command to execute."
}
},
"required": ["command"]
}
}
}
]
North Mini Code was evaluated on agentic coding and terminal benchmarks such as SWE-Bench and Terminal-Bench-style tasks. That makes it a better fit for repo-level workflows than a generic code model that only saw single-turn programming prompts.