multi_ai_cli.ini.example

# ==============================================================================
# Multi-AI CLI Configuration File
# New Agent / Engine Separated Layout
# ==============================================================================

[API_KEYS]
# API keys may also be provided via environment variables.
# Environment variables should override these values when present.
GEMINI_API_KEY = ...
OPENAI_API_KEY = ...
anthropic_api_key = ...
grok_api_key = ...
figma_access_token = ...

# ------------------------------------------------------------------------------
# Reusable model aliases
# This section is a catalog, not a single default slot per provider.
# Multiple models from the same provider may coexist.
# ------------------------------------------------------------------------------
[MODELS]
# --- Gemini family ---
gemini_pro         = gemini-3.1-pro-preview
gemini_flash       = gemini-3.1-flash
gemini_flash_image = gemini-3.1-flash-image-preview

# --- OpenAI family ---
gpt_main   = gpt-5.4
gpt_budget = gpt-4o-mini

# --- Claude family ---
claude_main   = claude-opus-4-6
claude_budget = claude-sonnet-4

# --- Grok family ---
grok_main = grok-4-latest

# --- Local models ---
local_qwen_coder = qwen2.5-coder:14b
local_qwen_chat  = qwen2.5:14b

# ------------------------------------------------------------------------------
# Global runtime behavior
# These are application/runtime controls, not model definitions.
# ------------------------------------------------------------------------------
[RUNTIME]
# Maximum number of conversation turns (user + assistant pairs) to keep per agent.
max_history_turns = 30

# Auto-continue tuning
auto_continue_max_rounds = 5
auto_continue_tail_chars = 1200

# ------------------------------------------------------------------------------
# Engine definitions
# Engine = physical execution backend
# Each engine binds provider type + model + credential/base_url + limits.
# ------------------------------------------------------------------------------

[ENGINE.openai_main]
type = openai
api_key_ref = OPENAI_API_KEY
model_ref = gpt_main
max_output_tokens = 4096

[ENGINE.openai_budget]
type = openai
api_key_ref = OPENAI_API_KEY
model_ref = gpt_budget
max_output_tokens = 4096

[ENGINE.gemini_main]
type = gemini
api_key_ref = GEMINI_API_KEY
model_ref = gemini_pro
max_output_tokens = 8192

[ENGINE.gemini_fast]
type = gemini
api_key_ref = GEMINI_API_KEY
model_ref = gemini_flash
max_output_tokens = 8192

[ENGINE.gemini_image]
type = gemini
api_key_ref = GEMINI_API_KEY
model_ref = gemini_flash_image
max_output_tokens = 8192

[ENGINE.claude_main]
type = anthropic
api_key_ref = anthropic_api_key
model_ref = claude_main
max_output_tokens = 8192

[ENGINE.claude_budget]
type = anthropic
api_key_ref = anthropic_api_key
model_ref = claude_budget
max_output_tokens = 8192

[ENGINE.grok_main]
type = grok
api_key_ref = grok_api_key
model_ref = grok_main
max_output_tokens = 8192

# Local OpenAI-compatible backend (Ollama / LM Studio / vLLM etc.)
[ENGINE.local_coder]
type = local_openai
base_url = http://localhost:11434/v1
api_key = ollama
model_ref = local_qwen_coder
max_output_tokens = 8192

[ENGINE.local_chat]
type = local_openai
base_url = http://localhost:11434/v1
api_key = ollama
model_ref = local_qwen_chat
max_output_tokens = 8192

# ------------------------------------------------------------------------------
# Agent definitions
# Agent = logical stateful actor
# Keep these thin: only logical identity -> engine binding
# ------------------------------------------------------------------------------

# OpenAI-family logical agents
[AGENT.gpt]
engine = openai_main

[AGENT.gpt.code]
engine = openai_main

[AGENT.gpt.review]
engine = openai_budget

[AGENT.gpt.plan]
engine = openai_main

[AGENT.gpt.chat]
engine = openai_budget

# Gemini-family logical agents
[AGENT.gemini]
engine = gemini_main

[AGENT.gemini.doc]
engine = gemini_main

[AGENT.gemini.chat]
engine = gemini_fast

[AGENT.gemini.image]
engine = gemini_image

# Claude-family logical agents
[AGENT.claude]
engine = claude_main

[AGENT.claude.plan]
engine = claude_main

[AGENT.claude.review]
engine = claude_budget

[AGENT.claude.doc]
engine = claude_main

# Grok-family logical agents
[AGENT.grok]
engine = grok_main

[AGENT.grok.chat]
engine = grok_main

# Local-family logical agents
[AGENT.local]
engine = local_chat

[AGENT.local.code]
engine = local_coder

[AGENT.local.test]
engine = local_coder

[AGENT.local.chat]
engine = local_chat

# ------------------------------------------------------------------------------
# Application paths
# ------------------------------------------------------------------------------
[Paths]
# Folder for prompt/persona assets used by commands like @efficient
work_efficient = prompts

# Blackboard directory for read/write artifacts (-r / -w)
work_data = work_data

# ------------------------------------------------------------------------------
# Logging
# ------------------------------------------------------------------------------
[logging]
enabled = true
log_dir = logs
base_filename = chat.log
max_bytes = 10485760
backup_count = 5
log_level = INFO

# ------------------------------------------------------------------------------
# Adapter / integration-specific configuration
# These are not engines.
# ------------------------------------------------------------------------------
[FIGMA]
handoff_dir = work_data/figma_handoff


# ------------------------------------------------------------------------------
# GitHub read-only adapter configuration
# Used by @github.* commands.
# Environment variables override these values:
#   GITHUB_TOKEN
#   GITHUB_API_BASE_URL
# For organization/private repositories, ensure the token Resource owner
# and repository permissions are configured correctly.
# ------------------------------------------------------------------------------
token = ...
api_base_url = https://api.github.com