AI Generators Overview

This page provides a high-level survey of all AI generators supported by ai-commit: what they are, which category they fall into, how to select one at runtime, and the minimum configuration required for each.

For a deeper look at how generators are implemented internally, see the Generator System Architecture (page 5) and its subpages. For the full set of config keys available per generator, see Generator-Specific Configuration (page 7.4).

---

Generator Categories

ai-commit supports two categories of generators:

Category	How it works	Generators
API-based	Sends the prompt via HTTP to a remote AI API	openai, openai_chat, moonshot, ernie_bot, ernie_bot_turbo
CLI-based	Shells out to a locally installed CLI tool	bito_cli, github_copilot_cli, github_models_cli

API-based generators use concrete HTTP client classes and require a valid API key. CLI-based generators use Symfony\Component\Process to invoke an external binary and require that binary to be installed and accessible on $PATH.

Sources: config/ai-commit.php78-219 README.md22-31

---

Generator Resolution

The GeneratorManager class (app/GeneratorManager.php) resolves a generator by name. It reads the generator configuration block from ai-commit.generators.<name>, converts the driver key to a studly-cased class name, and instantiates it.

Driver name → class name resolution (from createDriver in app/GeneratorManager.php63-77):

"openai" → App\Generators\OpenAIGenerator
"openai_chat" → App\Generators\OpenAIChatGenerator
"moonshot" → App\Generators\MoonshotGenerator
"ernie_bot" → App\Generators\ErnieBotGenerator
"ernie_bot_turbo" → App\Generators\ErnieBotTurboGenerator
"bito_cli" → App\Generators\BitoCliGenerator
"github_copilot_cli" → App\Generators\GithubCopilotCliGenerator
"github_models_cli" → App\Generators\GithubModelsCliGenerator

The string openai is specially handled before studly-casing to produce OpenAI instead of Openai.

Driver resolution diagram:

Sources: app/GeneratorManager.php41-77 app/Commands/CommitCommand.php72-78

---

Selecting a Generator

There are three ways to specify which generator to use:

1. At the command line (per-run):

2. As the persistent default in global config:

3. In a local config file (.ai-commit.json in the project directory):

The --generator option is defined in CommitCommand::configure() and defaults to the value of generator in config (app/Commands/CommitCommand.php150-156). Tab-completion for generator names is provided by CommitCommand::complete() (app/Commands/CommitCommand.php113-124).

Sources: app/Commands/CommitCommand.php132-190 config/ai-commit.php75

---

All Generators at a Glance

Relationship between config keys, classes, and external dependencies:

Sources: config/ai-commit.php78-219 app/GeneratorManager.php63-77

---

API-Based Generators

These generators communicate with a remote HTTP API. All API-based generators inherit from AbstractGenerator (app/Generators/AbstractGenerator.php) and use an HTTP client constructed from their config block.

openai — OpenAIGenerator

Uses the OpenAI Completions endpoint (/v1/completions). This is the older text-completion API. The default model is text-davinci-003.

Minimum configuration:

Config key	Default	Notes
api_key	sk-...	Required
parameters.model	text-davinci-003	Completions model
parameters.max_tokens	600
parameters.stream	true	Streaming response

Sources: config/ai-commit.php178-198

---

openai_chat — OpenAIChatGenerator

Uses the OpenAI Chat Completions endpoint (/v1/chat/completions). This is the default generator. The OpenAIChatGenerator class (app/Generators/OpenAIChatGenerator.php) extends OpenAIGenerator and overrides generate() to send a messages array instead of a plain prompt string.

Minimum configuration:

Config key	Default	Notes
api_key	sk-...	Required
parameters.model	gpt-3.5-turbo	Chat model
parameters.max_tokens	600
parameters.stream	true	Streaming response

Sources: config/ai-commit.php199-218 app/Generators/OpenAIChatGenerator.php

---

moonshot — MoonshotGenerator

Uses the Moonshot AI API, which is compatible with the OpenAI Chat Completions format. Requires a Moonshot API key (sk-...).

Minimum configuration:

Config key	Default	Notes
api_key	sk-...	Required
parameters.model	moonshot-v1-8k	Also: moonshot-v1-32k, moonshot-v1-128k
parameters.max_tokens	600
parameters.stream	true

Sources: config/ai-commit.php159-177

---

ernie_bot — ErnieBotGenerator

Uses Baidu's ERNIE-Bot large language model API. Requires both an API key and a secret key to obtain an OAuth access token.

Minimum configuration:

Config key	Default	Notes
api_key	...	Required
secret_key	...	Required — used for OAuth token exchange
parameters.temperature	0.95
parameters.stream	true

Sources: config/ai-commit.php103-115

---

ernie_bot_turbo — ErnieBotTurboGenerator

Identical configuration structure to ernie_bot, but targets the ERNIE-Bot-turbo model endpoint, which is faster and less expensive.

Minimum configuration:

Config key	Default	Notes
api_key	...	Required
secret_key	...	Required
parameters.stream	true

Sources: config/ai-commit.php116-128

---

CLI-Based Generators

These generators shell out to an external binary using Symfony\Component\Process. They inherit mustRunProcess and runProcess from AbstractGenerator (app/Generators/AbstractGenerator.php42-70). No HTTP client is used.

bito_cli — BitoCliGenerator

Invokes the Bito CLI (bito binary). The prompt is written to a temporary file, then passed via stdin or a prompt-file argument. The binary path is configurable.

Minimum configuration:

The bito binary must be installed. If it is not on $PATH, specify the path:

Config key	Default	Notes
binary	bito	Path or name of the bito binary
prompt_filename	bito.prompt	Temporary prompt file name
parameters.timeout	120	Process timeout in seconds
options	(various)	Extra CLI flags passed to bito

Sources: config/ai-commit.php79-102

---

github_copilot_cli — GithubCopilotCliGenerator

Invokes the GitHub Copilot CLI extension via the gh binary (gh copilot explain or similar subcommand). Requires the gh CLI with the Copilot extension installed and authenticated.

Minimum configuration:

The gh binary must be installed and authenticated. If it is not on $PATH:

Config key	Default	Notes
binary	gh	Path or name of the gh binary
parameters.timeout	120	Process timeout in seconds
options	--hostname: null	Extra CLI flags

Sources: config/ai-commit.php129-141

---

github_models_cli — GithubModelsCliGenerator

Invokes the GitHub Models CLI extension via the gh binary (gh models run). Requires the gh CLI with the Models extension installed and authenticated.

Minimum configuration:

Config key	Default	Notes
binary	gh	Path or name of the gh binary
model	openai/gpt-4.1-mini	Model identifier passed to gh models run
parameters.timeout	120	Process timeout in seconds
options['--system-prompt']	You are a git commit message generator.	System prompt injected as a flag
options['--max-tokens']	null
options['--temperature']	null

Sources: config/ai-commit.php142-158

---

Quick Configuration Reference

The table below summarizes the required setup step for each generator.

Config key	Generator class	Requires	Required env var (alternative)
openai	OpenAIGenerator	OpenAI API key	OPENAI_API_KEY
openai_chat	OpenAIChatGenerator	OpenAI API key	OPENAI_API_KEY
moonshot	MoonshotGenerator	Moonshot API key	MOONSHOT_API_KEY
ernie_bot	ErnieBotGenerator	Baidu API key + secret	ERNIE_API_KEY, ERNIE_SECRET_KEY
ernie_bot_turbo	ErnieBotTurboGenerator	Baidu API key + secret	ERNIE_API_KEY, ERNIE_SECRET_KEY
bito_cli	BitoCliGenerator	bito binary installed	BITO_CLI_BINARY
github_copilot_cli	GithubCopilotCliGenerator	gh + Copilot ext	GITHUB_COPILOT_CLI_BINARY
github_models_cli	GithubModelsCliGenerator	gh + Models ext	GITHUB_MODELS_CLI_BINARY

API keys can be set via environment variables (read by env() in config/ai-commit.php) or via the config set command to persist them in your global config file.

Sources: config/ai-commit.php78-219 README.md60-72

---

Default Generator

The default generator is openai_chat, as defined in config/ai-commit.php75:

This can be overridden globally:

GeneratorManager::getDefaultDriver() reads this value from the ai-commit.generator config key (app/GeneratorManager.php41-44).

Sources: config/ai-commit.php74-76 app/GeneratorManager.php41-44