05-model-packs-and-provider-strategy.md

layout	default
title	Chapter 5: Model Packs and Provider Strategy
nav_order	5
parent	Plandex Tutorial

Chapter 5: Model Packs and Provider Strategy

Welcome to Chapter 5: Model Packs and Provider Strategy. In this part of Plandex Tutorial: Large-Task AI Coding Agent Workflows, you will build an intuitive mental model first, then move into concrete implementation details and practical production tradeoffs.

Plandex supports combining models across providers to optimize quality, speed, and cost.

Strategy Tips

• use curated model packs as baseline
• test provider combinations on representative tasks
• track cost and latency by task class

Summary

You now have a model strategy framework for production Plandex usage.

Next: Chapter 6: Autonomy, Control, and Debugging

Source Code Walkthrough

app/shared/plan_model_settings.go

The GetWholeFileBuilderMaxTokens function in app/shared/plan_model_settings.go handles a key part of this chapter's functionality:

}

func (ps PlanSettings) GetWholeFileBuilderMaxTokens() int {
	modelPack := ps.GetModelPack()
	builder := modelPack.GetWholeFileBuilder()
	fallback := builder.GetFinalLargeContextFallback()
	return fallback.GetSharedBaseConfig(&ps).MaxTokens
}

func (ps PlanSettings) GetWholeFileBuilderMaxReservedOutputTokens() int {
	modelPack := ps.GetModelPack()
	builder := modelPack.GetWholeFileBuilder()
	fallback := builder.GetFinalLargeOutputFallback()
	return fallback.GetReservedOutputTokens(ps.CustomModelsById)
}

func (ps PlanSettings) GetPlannerMaxConvoTokens() int {
	modelPack := ps.GetModelPack()

	// for max convo tokens, we use the planner's default max convo tokens, *not* the fallback, so that we don't end up switching to the fallback just based on the conversation length
	planner := modelPack.Planner
	if planner.MaxConvoTokens != 0 {
		return planner.MaxConvoTokens
	}

	return planner.GetSharedBaseConfig(&ps).DefaultMaxConvoTokens
}

func (ps PlanSettings) GetPlannerEffectiveMaxTokens() int {
	maxPlannerTokens := ps.GetPlannerMaxTokens()
	maxReservedOutputTokens := ps.GetPlannerMaxReservedOutputTokens()

This function is important because it defines how Plandex Tutorial: Large-Task AI Coding Agent Workflows implements the patterns covered in this chapter.

app/shared/plan_model_settings.go

The GetWholeFileBuilderMaxReservedOutputTokens function in app/shared/plan_model_settings.go handles a key part of this chapter's functionality:

}

func (ps PlanSettings) GetWholeFileBuilderMaxReservedOutputTokens() int {
	modelPack := ps.GetModelPack()
	builder := modelPack.GetWholeFileBuilder()
	fallback := builder.GetFinalLargeOutputFallback()
	return fallback.GetReservedOutputTokens(ps.CustomModelsById)
}

func (ps PlanSettings) GetPlannerMaxConvoTokens() int {
	modelPack := ps.GetModelPack()

	// for max convo tokens, we use the planner's default max convo tokens, *not* the fallback, so that we don't end up switching to the fallback just based on the conversation length
	planner := modelPack.Planner
	if planner.MaxConvoTokens != 0 {
		return planner.MaxConvoTokens
	}

	return planner.GetSharedBaseConfig(&ps).DefaultMaxConvoTokens
}

func (ps PlanSettings) GetPlannerEffectiveMaxTokens() int {
	maxPlannerTokens := ps.GetPlannerMaxTokens()
	maxReservedOutputTokens := ps.GetPlannerMaxReservedOutputTokens()

	return maxPlannerTokens - maxReservedOutputTokens
}

func (ps PlanSettings) GetArchitectEffectiveMaxTokens() int {
	maxArchitectTokens := ps.GetArchitectMaxTokens()
	maxReservedOutputTokens := ps.GetArchitectMaxReservedOutputTokens()

This function is important because it defines how Plandex Tutorial: Large-Task AI Coding Agent Workflows implements the patterns covered in this chapter.

app/shared/plan_model_settings.go

The GetPlannerMaxConvoTokens function in app/shared/plan_model_settings.go handles a key part of this chapter's functionality:

}

func (ps PlanSettings) GetPlannerMaxConvoTokens() int {
	modelPack := ps.GetModelPack()

	// for max convo tokens, we use the planner's default max convo tokens, *not* the fallback, so that we don't end up switching to the fallback just based on the conversation length
	planner := modelPack.Planner
	if planner.MaxConvoTokens != 0 {
		return planner.MaxConvoTokens
	}

	return planner.GetSharedBaseConfig(&ps).DefaultMaxConvoTokens
}

func (ps PlanSettings) GetPlannerEffectiveMaxTokens() int {
	maxPlannerTokens := ps.GetPlannerMaxTokens()
	maxReservedOutputTokens := ps.GetPlannerMaxReservedOutputTokens()

	return maxPlannerTokens - maxReservedOutputTokens
}

func (ps PlanSettings) GetArchitectEffectiveMaxTokens() int {
	maxArchitectTokens := ps.GetArchitectMaxTokens()
	maxReservedOutputTokens := ps.GetArchitectMaxReservedOutputTokens()

	return maxArchitectTokens - maxReservedOutputTokens
}

func (ps PlanSettings) GetCoderEffectiveMaxTokens() int {
	maxCoderTokens := ps.GetCoderMaxTokens()
	maxReservedOutputTokens := ps.GetCoderMaxReservedOutputTokens()

This function is important because it defines how Plandex Tutorial: Large-Task AI Coding Agent Workflows implements the patterns covered in this chapter.

app/shared/plan_model_settings.go

The GetPlannerEffectiveMaxTokens function in app/shared/plan_model_settings.go handles a key part of this chapter's functionality:

}

func (ps PlanSettings) GetPlannerEffectiveMaxTokens() int {
	maxPlannerTokens := ps.GetPlannerMaxTokens()
	maxReservedOutputTokens := ps.GetPlannerMaxReservedOutputTokens()

	return maxPlannerTokens - maxReservedOutputTokens
}

func (ps PlanSettings) GetArchitectEffectiveMaxTokens() int {
	maxArchitectTokens := ps.GetArchitectMaxTokens()
	maxReservedOutputTokens := ps.GetArchitectMaxReservedOutputTokens()

	return maxArchitectTokens - maxReservedOutputTokens
}

func (ps PlanSettings) GetCoderEffectiveMaxTokens() int {
	maxCoderTokens := ps.GetCoderMaxTokens()
	maxReservedOutputTokens := ps.GetCoderMaxReservedOutputTokens()

	return maxCoderTokens - maxReservedOutputTokens
}

func (ps PlanSettings) GetWholeFileBuilderEffectiveMaxTokens() int {
	maxWholeFileBuilderTokens := ps.GetWholeFileBuilderMaxTokens()
	maxReservedOutputTokens := ps.GetWholeFileBuilderMaxReservedOutputTokens()

	return maxWholeFileBuilderTokens - maxReservedOutputTokens
}

func (ps PlanSettings) GetModelProviderOptions() ModelProviderOptions {
	opts := ModelProviderOptions{}

This function is important because it defines how Plandex Tutorial: Large-Task AI Coding Agent Workflows implements the patterns covered in this chapter.

How These Components Connect

flowchart TD
    A[GetWholeFileBuilderMaxTokens]
    B[GetWholeFileBuilderMaxReservedOutputTokens]
    C[GetPlannerMaxConvoTokens]
    D[GetPlannerEffectiveMaxTokens]
    E[GetArchitectEffectiveMaxTokens]
    A --> B
    B --> C
    C --> D
    D --> E

Loading