Profile Management: Multiple Configurations per Backend

GPTCode now supports profiles - multiple named configurations per backend. This lets you quickly switch between different model setups without editing configuration files.

Why Profiles?

Different tasks need different model configurations:

Speed profile: Fast models for all agents (lower latency, lower cost)
Quality profile: Best models for complex tasks (higher accuracy)
Free profile: Only free-tier models for experimentation
Production profile: Reliable, battle-tested models

Instead of manually editing ~/.gptcode/setup.yaml, create and switch between profiles.

Quick Start

Show Current Profile

gt profile
# Current: groq/default
#   router: llama-3.1-8b-instant
#   query: llama-3.3-70b-versatile
#   editor: moonshotai/kimi-k2-instruct-0905
#   research: groq/compound

List Available Profiles

gt profile list
# groq.default (current)
# groq.speed
# groq.quality
# openrouter.default
# openrouter.free

gt profile list groq
# groq.default (current)
# groq.speed
# groq.quality

Show Profile Configuration

gt profile show groq.speed
# groq/speed
#   router: llama-3.1-8b-instant
#   query: llama-3.1-8b-instant
#   editor: llama-3.1-8b-instant
#   research: llama-3.1-8b-instant

Switch Profile

gt profile use groq.speed
# ✓ Switched to groq/speed

Create New Profile

gt profiles create groq speed
# ✓ Created profile: groq/speed
# 
# Configure agent models using:
#   gt profiles set-agent groq speed router <model>
#   gt profiles set-agent groq speed query <model>
#   gt profiles set-agent groq speed editor <model>
#   gt profiles set-agent groq speed research <model>

Configure Agent Models

gt profiles set-agent groq speed router llama-3.1-8b-instant
gt profiles set-agent groq speed query llama-3.1-8b-instant
gt profiles set-agent groq speed editor llama-3.1-8b-instant
gt profiles set-agent groq speed research llama-3.1-8b-instant
# ✓ Set groq/speed router = llama-3.1-8b-instant
# ...

Profile Structure in setup.yaml

Profiles are stored in your ~/.gptcode/setup.yaml:

defaults:
    backend: groq
    profile: default  # currently active profile

backend:
    groq:
        # ... backend config ...
        agent_models:     # "default" profile (backwards compatible)
            router: llama-3.1-8b-instant
            query: llama-3.3-70b-versatile
            editor: moonshotai/kimi-k2-instruct-0905
            research: groq/compound
        profiles:         # named profiles
            speed:
                agent_models:
                    router: llama-3.1-8b-instant
                    query: llama-3.1-8b-instant
                    editor: llama-3.1-8b-instant
                    research: llama-3.1-8b-instant
            quality:
                agent_models:
                    router: llama-3.3-70b-versatile
                    query: llama-3.3-70b-versatile
                    editor: llama-3.3-70b-versatile
                    research: groq/compound

Switching Profiles in Neovim

In Neovim, press Ctrl+M to open the model selector:

Select backend (e.g., “groq”)
Select profile (e.g., “speed”, “quality”)
The chat header updates to show active profile

🐺 GPTCode
Backend: Groq / speed
  router: llama-3.1-8b-instant
  query: llama-3.1-8b-instant
  editor: llama-3.1-8b-instant
  research: llama-3.1-8b-instant

Example: OpenRouter Free Models

Create a profile using only free-tier OpenRouter models:

gt profiles create openrouter free

gt profiles set-agent openrouter free router \
  google/gemini-2.0-flash-exp:free

gt profiles set-agent openrouter free query \
  google/gemini-2.0-flash-exp:free

gt profiles set-agent openrouter free editor \
  moonshotai/kimi-k2:free

gt profiles set-agent openrouter free research \
  google/gemini-2.0-flash-exp:free

Now you can experiment with free models without API costs.

Agent Types

Each profile configures four agent types:

router: Fast model for intent classification (determines which agent handles request)
query: Model for reading and analyzing code
editor: Model for writing and modifying code
research: Model for web search and documentation lookup

Best Practices

Speed Profile

Use fast, cheap models for rapid iteration:

gt profiles create groq speed
gt profiles set-agent groq speed router llama-3.1-8b-instant
gt profiles set-agent groq speed query llama-3.1-8b-instant
gt profiles set-agent groq speed editor llama-3.1-8b-instant
gt profiles set-agent groq speed research llama-3.1-8b-instant

Quality Profile

Use best available models for complex tasks:

gt profiles create groq quality
gt profiles set-agent groq quality router llama-3.1-8b-instant  # routing is simple
gt profiles set-agent groq quality query llama-3.3-70b-versatile
gt profiles set-agent groq quality editor llama-3.3-70b-versatile
gt profiles set-agent groq quality research groq/compound

Specialized Profiles

Create profiles for specific use cases:

# Code-heavy tasks
gt profiles create groq coding
gt profiles set-agent groq coding editor deepseek-v3

# Research-heavy tasks
gt profiles create groq docs
gt profiles set-agent groq docs research groq/compound

Migration from Old Config

If you have agent_models at the backend root level (old format), it automatically becomes the “default” profile. No manual migration needed.

Old format:

backend:
    groq:
        agent_models:
            router: ...

Works as:

gt profile show groq.default
# Shows the models from agent_models

Troubleshooting

Profile Not Found

gt profile list groq
# Check if profile exists

gt profiles create groq myprofile
# Create if missing (use plural 'profiles' for creation)

Wrong Models Showing

Verify profile configuration:

gt profile show groq.myprofile
# Check each agent's model

gt profiles set-agent groq myprofile router correct-model
# Fix individual agents (use plural 'profiles' for configuration)

Profile Changes Not Reflecting in Neovim

Restart Neovim or reload the buffer. The plugin reads configuration on startup.

Implementation Details

Profiles use proper YAML marshaling (not manual parsing), ensuring:

Clean, maintainable code
Proper error handling
Type safety
Easy extensibility

The profile system replaces fragile space-counting logic with Go’s yaml.v3 library.

Future Enhancements

Planned features:

gt profiles copy <src> <dst> - Clone existing profile
gt profiles delete <backend> <profile> - Remove profile
gt profiles export/import - Share profiles between machines
Profile templates for common use cases

Profiles make it easy to switch between different model configurations without manual editing. Create profiles for different use cases and switch seamlessly in Neovim or via CLI.