Ask

A command-line interface for interacting with AI language models. Currently supports Anthropic's Claude, with more providers planned.

Disclaimer

This file has been AI edited and human verified. No other files in this project have been touched by AI.

Quick Start

# Install and setup
crystal build ask.cr

# Create a new conversation
mkdir -p ai/weather
cd ai/weather
echo "Tell me about the hottest place on earth." > question
ask
cat answer

Installation

1. Build the binary:

   crystal build ask.cr

2. Configure credentials in ~/.creds.yml:

   # put keys here for whatever engines you have access to
   claude:
     api_key: "your_api_key_here"
   gemini:
     api_key: "your_api_key_here"

3. Configure settings in ~/.ask.yml:

   default_model: "claude-3-5-sonnet-latest"
   # or "gemini-1.5-pro
   providers:
     claude: {}
     gemini: {}

Usage

Basic Commands

• Start a new conversation: Create a directory and add your question
• Ask a question: Write to question file and run ask
• View response: Read answer file (symlinked to latest response)
• Follow up: Update question file and run ask again

File Structure

• Questions are saved as alternating sequential files ending in q (e.g., 01q, 03q)
• Answers are saved as alternating sequential files ending in a (e.g., 02a, 04a)
• Latest answer is always symlinked to answer

Attachments

• Lines like <<filename will attach filename to the question. filename can be absolute, or relative to the current directory.
• These lines are not sent to the AI.
• After the first successful use, attachments are copied to the current directory for reuse and later reference.
• Attachments are named Xq.bY where X is your question's sequence number and Y is your attachment's sequence number for that question.

Advanced Features

Model Aliases

You can define aliases for models in your ~/.ask.yml file:

providers:
  anthropic:
    models:
      claude-3-7-sonnet-latest:
        - s37
  google:
    models:
      gemini-2.5-pro-exp-03-25:
        - g25

Then, within a conversation directory, setting .model to s37 will use claude-3-7-sonnet-latest and g25 will use gemini-2.5-pro-exp-03-25. For example:

echo "s37" > .model
ask

Tool Calling

ask allows you to use external tools and incorporate their results into the conversation. All tool runs are logged in ~/.ask/logs/<name>/<date>/<time>. See sample-tools for examples.

1. Tool Directory: Place your executable tools in the ~/.ask/tools directory. Each tool in ~/.ask/tools must be in its own directory.

   Each tool directory must include:

   • exe: The tool itself. Permissions must include u+x. Reads JSON from standard input and writes JSON to standard output.
   • description: A plain text file describing the tool's inputs, outputs, and purpose. Sent to the AI as documentation.
   • schema: A JSON schema describing the tool's inputs.
   • provider: A file containing either a single asterisk or the name of a provider. Needed for future work on provider-provided internal tools.

2. List Tools: Use ask tool to see a list of available tools detected in the ~/.ask/tools directory.

3. Enable Tool: To enable a tool for the current conversation, use the ask tool enable <tool_name> command within the conversation directory. For example:

   cd ai/weather # or wherever you are working
   ask tool enable add

4. Using Enabled Tools: When the AI determines that a tool needs to be used, it will request its use. You will be prompted with the tool name and arguments proposed by the AI. For example:

   echo "What is 4 + 4?" > question
   ask

   ask will output something like:

   add {"a": 4, "b": 4}
   

   Press <enter> to run the tool or ctrl-c to exit. For brevity, no prompt is displayed.

   • Security Note: Each tool execution requires manual approval for security reasons, preventing potentially harmful or unintended actions.

5. Tool Output: If approved, the tool's output will be sent back to the AI, and the conversation will continue. This cycle will repeat as needed until:

   • The AI no longer requests tool use.
   • The user ends the conversation.
   • The user aborts the session by pressing <ctrl-c> at the tool request prompt.

Conversation Management

• Restart Conversation: ask restart

  • Creates a fresh start while preserving history
  • Previous Q&A won't be sent to AI in future interactions

• Remove Content: Delete or move Q&A pairs to exclude them from context

Configuration

• Disable Prompt Caching:

  echo "" > .nocache

• Change Model:

  echo "model-name" > .model

Extended Example

# Start conversation about weather
mkdir -p ai/weather
cd ai/weather
echo "What is the coldest city on earth?" > question
ask
cat answer

# Follow up
echo "Tell me more about the place(s) you mentioned above." > question
ask

# Start new topic
ask restart
echo "Tell me how to cook pasta." > question
ask

# Organize conversations
mkdir ../cooking
mv 03q 04a ../cooking/

# Re-enable full history
rm .restart

# Continue previous conversation
echo "Tell me even more about the above places." > question
ask

# ask about an image from the web
mkdir ../web-img
cd web-img
wget "https://example.com/image.png"
echo '<<image.png' > question
echo "Describe this image." >> question
ask

Contributing

Feel free to open issues for:

• Bug reports
• Feature requests
• New AI provider suggestions (please include a link to API documentation if possible)

Support

Currently supported providers:

• Anthropic Claude
• Google Gemini

More providers planned - open an issue to request support for additional AI models.