OCCode - New Features Guide v0.2.0

Model Profiles, Convergence Engine, and Multi-Model Management

📋 Feature Overview

OCCode now includes three powerful new features for managing and optimizing AI model usage:

📚 Model Catalog

The Model Catalog is a centralized registry that simplifies working with multiple AI providers and models.

Supported Providers

Provider	Models Available	Environment Variable
Anthropic	Claude Opus 4, Sonnet 4, Haiku 4.5	ANTHROPIC_API_KEY
OpenAI	GPT-4o, GPT-4, o1, o3-mini, GPT-4o-mini	OPENAI_API_KEY
Google	Gemini 2.0 Flash, Gemini 2.5 Pro	GOOGLE_API_KEY
DeepSeek	DeepSeek V3, DeepSeek R1	DEEPSEEK_API_KEY
Mistral	Mistral Large, Codestral	MISTRAL_API_KEY
Groq	Llama 3.3 70B, DeepSeek R1 Distill	GROQ_API_KEY
Together	Various open models	TOGETHER_API_KEY
OpenRouter	100+ models via unified API	OPENROUTER_API_KEY
OpenCan	Custom models	OPENCAN_API_KEY
Ollama	Local models (CodeLlama, Qwen, etc.)	None (local)
LM Studio	Local models	None (local)

Short Aliases

Use convenient short names instead of full model identifiers:

Checking Available Models

# See all models in catalog
/converge catalog

# See only models with API keys set
/converge available

# Search for specific capabilities
/converge search code
/converge search fast
/converge search free

👤 Model Profiles

Model Profiles let you create named configurations for different AI models and switch between them instantly.

Creating Profiles

From Built-in Templates

# List available templates
/profile templates

# Create from template
/profile template fast
/profile template power
/profile template creative

Interactive Creation

# Create custom profile
/profile create myprofile

# You'll be prompted for:
# - Provider (anthropic, openai, etc.)
# - Model (sonnet, gpt4o, etc.)
# - Temperature (0.0-1.0)
# - Max tokens
# - Description
# - Prefix trigger (optional)

Built-in Templates

Template	Model	Prefix	Use Case
fast	Claude Haiku	quick:	Quick questions, simple tasks
default	Claude Sonnet 4	-	Balanced quality/speed
power	Claude Opus 4	think:	Complex reasoning, critical tasks
creative	Sonnet (high temp)	create:	Creative writing, brainstorming
gpt	GPT-4 Turbo	gpt:	Alternative perspective
local	Ollama	local:	Privacy, offline work

Using Profiles

Activate a Profile

# Switch to a profile
/profile fast
/profile power
/profile myprofile

# View current profile and all available
/profile

# Deactivate profile (return to global config)
/profile off

Prefix Shortcuts

Set a prefix trigger to temporarily use a profile for a single response:

# Set prefix for a profile
/profile prefix fast quick:
/profile prefix power think:

# Use prefix in conversation
quick: what is recursion?
think: design a distributed cache system

Managing Profiles

# Edit existing profile
/profile edit myprofile

# Delete profile
/profile delete myprofile

# Set default profile (auto-activates on startup)
/profile default fast

# Export profiles (for sharing with team)
/profile export

# Import profiles
/profile import profiles.json

Profile Configuration

Profiles are stored in ~/.occode/profiles.json with secure file permissions (0600):

{
  "profiles": {
    "fast": {
      "name": "fast",
      "provider": "anthropic",
      "model": "claude-haiku-4-5-20251001",
      "temperature": 0.3,
      "maxTokens": 2048,
      "triggerPrefix": "quick:",
      "description": "Quick answers, low cost",
      "createdAt": "2025-02-01T..."
    }
  },
  "activeProfile": "fast",
  "defaultProfile": null
}

⚡ Convergence Engine

The Convergence Engine runs multiple AI models in parallel and combines their outputs for higher quality results.

Core Concept

Instead of asking one model for an answer, convergence:

1. Generates - Multiple models work on the task simultaneously
2. Synthesizes - An aggregator model combines the best parts
3. Returns - A single, higher-quality response

Convergence Strategies

Strategy	How It Works	Best For	API Calls
merge	All models generate → aggregator synthesizes	High-quality code, complex tasks	N + 1
vote	All models generate → all vote on best	Multiple valid solutions, picking best	N + N
debate	Generate → critique → revise (multiple rounds)	Critical decisions, thorough analysis	N × rounds × 3
review	One generates → another reviews → first revises	Code review workflow, cost-effective	3-4

Quick Start with Presets

The easiest way to use convergence is with built-in presets:

# View available presets
/converge preset

# Apply a preset
/converge preset duo-merge
/converge preset code-review
/converge preset trio-merge

# Enable convergence
/converge on

# Now your messages use convergence
write a binary search function

# Disable when done
/converge off

Built-in Presets

Preset	Models	Strategy	Description
duo-merge	Sonnet + GPT-4 → Sonnet	merge	Best balance of quality and cost
trio-merge	Sonnet + GPT-4 + Haiku → Sonnet	merge	Maximum quality, diverse perspectives
code-review	Sonnet → GPT-4 (review)	review	Cost-effective quality assurance
debate	Sonnet ↔ GPT-4 (2 rounds)	debate	Thorough analysis through dialogue
vote	3 models generate + vote	vote	Democratic selection
local-merge	CodeLlama + DeepSeek (local)	merge	Private, no API costs

Custom Convergence Configuration

Adding Models

# Add models using short aliases
/converge add sonnet
/converge add gpt4o
/converge add haiku

# Or specify explicitly
/converge add mymodel anthropic claude-sonnet-4-20250514

# Remove models
/converge remove haiku

# View configured models
/converge models

Setting Strategy

# Change strategy
/converge strategy merge
/converge strategy vote
/converge strategy debate
/converge strategy review

# View strategy details
/converge strategy

Configuring Aggregator

# Set which model synthesizes results (merge strategy)
/converge aggregator sonnet
/converge aggregator gpt4o

# View current aggregator
/converge aggregator

Debate Rounds

# Set number of debate rounds (1-5)
/converge rounds 3

# View current setting
/converge rounds

Viewing Results

# Show individual model outputs (before synthesis)
/converge show on

# Only show final converged output
/converge show off

# View last convergence result with statistics
/converge last

# View current status and configuration
/converge

Configuration Management

# Export configuration
/converge export my-config.json

# Import configuration
/converge import my-config.json

# Reset to defaults
/converge reset

🧪 Testing Features

Testing Model Profiles

Test 1: Create and Activate Profile

# 1. Create a fast profile
/profile template fast

# 2. Verify it was created
/profile

# 3. Activate it
/profile fast

# 4. Verify it's active (should show checkmark)
/profile

Expected Result: Profile is listed with a green → indicator showing it's active.

Test 2: Prefix Shortcuts

# 1. Ensure fast profile has prefix
/profile prefix fast quick:

# 2. Test prefix usage
quick: what is 2 + 2?

# 3. Send regular message
what is 3 + 3?

Expected Result:

• First message shows "Using profile 'fast'" notification
• Second message uses your regular active profile
• Both get responses

Test 3: Profile Switching

# 1. Create power profile
/profile template power

# 2. Switch to power
/profile power

# 3. Verify active
/profile

# 4. Switch back
/profile off

Expected Result: Status shows correct active profile at each step.

Testing Convergence Engine

Test 1: Basic Convergence

# 1. Load a preset
/converge preset duo-merge

# 2. Check configuration
/converge

# 3. Enable convergence
/converge on

# 4. Send a task
write a function to check if a number is prime

# 5. Disable
/converge off

Expected Result:

• Spinner shows "Converging [merge] with sonnet, gpt4o..."
• Response is generated
• Statistics shown: strategy, model count, time, cost

Test 2: View Individual Outputs

# 1. Enable individual output display
/converge show on

# 2. Enable convergence
/converge on

# 3. Send task
explain quicksort algorithm

# 4. Observe output

Expected Result:

• Individual outputs section shows each model's response (truncated)
• Converged output section shows final synthesis
• Statistics include all models

Test 3: Different Strategies

# Test Review Strategy
/converge preset code-review
/converge on
write a binary search function
/converge off

# Test Vote Strategy
/converge preset vote
/converge on
what's the best sorting algorithm for large datasets?
/converge off

# View last results
/converge last

Expected Result: Each strategy produces different execution patterns visible in the statistics.

Testing Model Catalog

# 1. View all available models
/converge available

# 2. Search for specific models
/converge search fast
/converge search code

# 3. Browse full catalog
/converge catalog

# 4. Filter by provider
/converge search anthropic

Expected Result: Models are listed with pricing, capabilities, and availability status (✓ or ✗ based on API key).

Integration Test: Profiles + Convergence

# 1. Create profile with specific model
/profile create test_profile
# Choose: anthropic, sonnet, 0.7, 4096

# 2. Activate it
/profile test_profile

# 3. Enable convergence
/converge preset duo-merge
/converge on

# 4. Send task
explain recursion

# 5. Check that convergence works with profile active
/converge
/profile

Expected Result: Convergence uses its configured models (not the active profile). Both systems work independently.

🔧 Troubleshooting

Model Profiles Issues

Profile Not Found

Issue: Profile "myprofile" not found

Solution:

# List all profiles to check spelling
/profile

# Recreate if needed
/profile template fast

Prefix Not Triggering

Issue: Typing quick: message doesn't activate profile

Solution:

• Check prefix is set: /profile (should show [quick:] next to profile)
• Ensure exact match with colon: quick: not quick
• Set prefix: /profile prefix fast quick:

API Key Missing for Profile

Issue: No API key for anthropic when activating profile

Solution:

# Set API key for the provider
export ANTHROPIC_API_KEY="sk-ant-..."

# Or set via OCCode
occode config --set-key --provider anthropic

Profile File Permissions

Issue: Cannot save profiles

Solution:

# Check and fix permissions
chmod 600 ~/.occode/profiles.json

# Recreate if corrupted
rm ~/.occode/profiles.json
/profile template fast

Convergence Engine Issues

No Models Configured

Issue: No models configured yet when enabling convergence

Solution:

# Use a preset
/converge preset duo-merge

# Or add models manually
/converge add sonnet
/converge add gpt4o
/converge aggregator sonnet

Model Not Available

Issue: Model sonnet not available

Solution:

# Check available models
/converge available

# Ensure API key is set
export ANTHROPIC_API_KEY="sk-ant-..."

# Verify in catalog
/converge catalog

Convergence Taking Too Long

Issue: Convergence requests timeout or take very long

Solution:

• Reduce models: Use duo-merge instead of trio-merge
• Change strategy: Review is faster than debate
• Check API limits: You might be hitting rate limits
• Network issues: Verify internet connection

High Costs

Issue: Convergence is expensive

Solution:

# Use cost-effective strategy
/converge preset code-review

# Use cheaper models
/converge add haiku
/converge add mini
/converge aggregator haiku

# Disable when not needed
/converge off

# Check cost before running
/converge  # Shows last run cost

Individual Outputs Not Showing

Issue: Can't see what each model generated

Solution:

# Enable individual output display
/converge show on

# View last convergence details
/converge last

Model Catalog Issues

Model Alias Not Recognized

Issue: Unknown model alias: xyz

Solution:

# Search catalog for correct alias
/converge search xyz

# View all available aliases
/converge catalog

# Use explicit form if needed
/converge add myname provider full-model-id

No Available Models

Issue: /converge available shows no models

Solution:

# Check environment variables
env | grep API_KEY

# Set missing API keys
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."

# Restart OCCode to pick up new env vars

General Troubleshooting

Features Not Available

Issue: /profile or /converge command not found

Solution:

# Verify OCCode version
occode --version
# Should be v0.2.0 or later

# Update by downloading the latest version from opencan.ai/downloads
# Then replace the binary:
tar xzf occode-latest-linux-x64.tar.gz
sudo mv occode /usr/local/bin/

Config File Corruption

Issue: Features behaving strangely or errors about JSON

Solution:

# Backup existing config
cp ~/.occode/profiles.json ~/.occode/profiles.json.bak
cp ~/.occode/convergence.json ~/.occode/convergence.json.bak

# Remove corrupted configs
rm ~/.occode/profiles.json
rm ~/.occode/convergence.json

# Recreate from scratch
/profile template fast
/converge preset duo-merge

Get More Help

# View built-in help
/help

# Check OCCode documentation
https://opencan.ai/docs

# Contact support
support@opencan.ai

# Community forum
https://opencan.ai/community

Quick Reference Card

Category	Command	Description
Profiles	/profile	Show status and list all
	/profile template fast	Create from template
	/profile <name>	Activate profile
	quick: message	Use prefix shortcut
Convergence	/converge preset duo-merge	Load preset
	/converge on	Enable convergence
	/converge off	Disable convergence
	/converge last	View last result stats
Catalog	/converge catalog	Browse all models
	/converge available	Show available models
	/converge search <term>	Search models