gitbash

gitbash is a terminal CLI for asking questions about GitHub repositories.

Install it globally:

npm i -g @invertase/gitbash

Then use it locally:

gitbash --repo invertase/react-native-firebase "How does auth work?"

Quick start

Set your defaults once:

gitbash config \
  --model anthropic/claude-haiku-4-5 \
  --anthropic-token "$ANTHROPIC_API_KEY"

Then ask questions:

gitbash --repo invertase/react-native-firebase "How does auth work?"
gitbash --repo invertase/react-native-firebase --ref next "Summarize the Firebase app setup"
gitbash "Tell me about invertase/react-native-firebase"

If --repo is omitted, gitbash will try to infer owner/repo from the prompt.

Models

--model supports provider-prefixed models such as:

• anthropic/claude-haiku-4-5
• openai/gpt-4.1
• google/gemini-2.5-flash
• xai/grok-4
• local or local/<model-id>: OpenAI-compatible servers on your machine (LM Studio, Ollama /v1, etc.)

Examples:

gitbash --model anthropic/claude-haiku-4-5 --repo invertase/react-native-firebase "Explain auth"
gitbash --model openai/gpt-4.1 --repo invertase/react-native-firebase "Explain auth"
gitbash --model google/gemini-2.5-flash --repo invertase/react-native-firebase "Explain auth"
gitbash --model xai/grok-4 --repo invertase/react-native-firebase "Explain auth"

Local LLM (example: LM Studio serving at http://localhost:1234/v1):

export GITBASH_LOCAL_BASE_URL=http://localhost:1234/v1
gitbash --model local --repo invertase/react-native-firebase "Summarize the README"
# or pin a model id your server exposes:
gitbash --model local/qwen2.5-coder-7b --local-base-url http://localhost:1234/v1 --repo invertase/react-native-firebase "Explain auth"

local uses the server default model; local/<id> names a specific model. An API key is optional for most local servers; use --local-token or GITBASH_LOCAL_API_KEY if yours requires one.

Bare Anthropic model names also work:

gitbash --model claude-haiku-4-5 --repo invertase/react-native-firebase "Explain auth"

That is treated as anthropic/claude-haiku-4-5.

Config

Configuration is stored in ~/.gitbash/config.json.

Set a default model and token:

gitbash config \
  --model google/gemini-2.5-flash \
  --google-token "$GOOGLE_GENERATIVE_AI_API_KEY"

Store multiple provider tokens:

gitbash config \
  --anthropic-token "$ANTHROPIC_API_KEY" \
  --openai-token "$OPENAI_API_KEY" \
  --google-token "$GOOGLE_GENERATIVE_AI_API_KEY" \
  --xai-token "$XAI_API_KEY"

Default local OpenAI-compatible base URL and model (optional --local-token if your server needs it):

gitbash config \
  --model local \
  --local-base-url http://localhost:1234/v1

Override config for one command:

gitbash \
  --model openai/gpt-4.1 \
  --openai-token "$OPENAI_API_KEY" \
  --repo invertase/react-native-firebase \
  "How does auth work?"

Flags

• --repo <owner/repo>: repository to inspect
• --ref <ref>: git ref to mount, defaults to main
• --model <provider/model>: model to use
• --token <token>: generic provider token
• --anthropic-token <token>: Anthropic API token
• --openai-token <token>: OpenAI API token
• --google-token <token>: Google API token
• --xai-token <token>: xAI API token
• --local-base-url <url>: OpenAI-compatible API base URL for local models (e.g. http://localhost:1234/v1)
• --local-token <token>: optional API key for local OpenAI-compatible servers
• --api-url <url>: override the gitbash API base URL

Environment variables

You can use environment variables instead of flags:

• GITBASH_MODEL
• GITBASH_API_URL
• GITBASH_API_TOKEN
• GITBASH_ANTHROPIC_API_KEY or ANTHROPIC_API_KEY
• GITBASH_OPENAI_API_KEY or OPENAI_API_KEY
• GITBASH_GOOGLE_API_KEY, GOOGLE_GENERATIVE_AI_API_KEY, or GOOGLE_API_KEY
• GITBASH_XAI_API_KEY or XAI_API_KEY
• GITBASH_LOCAL_BASE_URL: base URL for local models (no trailing slash)
• GITBASH_LOCAL_API_KEY: optional API key for local OpenAI-compatible servers

Resolution order:

• CLI flags
• environment variables
• ~/.gitbash/config.json
• built-in default for apiUrl

model must be set before a prompt can run. Cloud providers also need a matching API token; the local provider does not require a token unless your server enforces one.