coaiapy 0.2.56

A Python package for audio transcription, synthesis, and tagging using Boto3.

CoAiAPy

CoAiAPy is a Python package that provides functionality for audio transcription, synthesis, and tagging of MP3 files using Boto3 and the Mutagen library. This package is designed to facilitate the processing of audio files for various applications.

Features

• Audio Transcription: Convert audio files to text using AWS services.
• Audio Synthesis: Generate audio files from text input.
• MP3 Tagging: Add metadata tags to MP3 files for better organization and identification.
• Redis Stashing: Stash key-value pairs to a Redis service.

Installation

To install the package, you can use pip:

pip install coaiapy

Usage

CLI Tool

CoAiAPy provides a CLI tool for audio transcription, summarization, and stashing to Redis.

Help

To see the available commands and options, use the --help flag:

coaia --help

Setup

Set these environment variables to use the AWS transcription service:

OPENAI_API_KEY
AWS_KEY_ID
AWS_SECRET_KEY
AWS_REGION
REDIS_HOST
REDIS_PORT
REDIS_PASSWORD
REDIS_SSL

Transcribe Audio

To transcribe an audio file to text:

coaia transcribe <file_path>

Example:

coaia transcribe path/to/audio/file.mp3

Summarize Text

To summarize a text:

coaia summarize <text>

Example:

coaia summarize "This is a long text that needs to be summarized."

To summarize text from a file:

coaia summarize --f <file_path>

Example:

coaia summarize --f path/to/text/file.txt

Stash Key-Value Pair to Redis

To stash a key-value pair to Redis:

coaia tash <key> <value>

Example:

coaia tash my_key "This is the value to stash."

To stash a key-value pair from a file:

coaia tash <key> --f <file_path>

Example:

coaia tash my_key --f path/to/value/file.txt

Fetch Value from Redis

To fetch a value from Redis by key:

coaia fetch <key>

Example:

coaia fetch my_key

To fetch a value from Redis and save it to a file:

coaia fetch <key> --output <file_path>

Example:

coaia fetch my_key --output path/to/output/file.txt

Process Custom Tags

Enable custom quick addons for assistants or bots using process tags. To add a new process tag to coaia.json, include entries like:

	"dictkore_temperature":0.2,
	"dictkore_instruction": "You do : Receive a dictated text that requires correction and clarification.\n\n# Corrections\n\n- In the dictated text, spoken corrections are made. You make them and remove the text related to that to keep the essence of what is discussed.\n\n# Output\n\n- You keep all the essence of the text (same length).\n- You keep the same style.\n- You ensure annotated dictation errors in the text are fixed.",

coaia p dictkore "my text to correct"

Building and Publishing

Use the provided Makefile to build and distribute the package. Typical tasks:

make build        # create sdist and wheel
make dist         # alias for make build
make upload-test  # upload the distribution to Test PyPI
make test-release # bump patch version, clean, build, and upload to Test PyPI

Both upload tasks use: twine upload --repository testpypi dist/* make test-release automatically sources $HOME/.env so TWINE_USERNAME and TWINE_PASSWORD are available. If you need the variables in your shell, run:

export $(grep -v '^#' $HOME/.env | xargs)

It also bumps the patch version using bump.py before uploading.

Langfuse Integration (fuse)

CoAiAPy integrates with Langfuse to manage prompts, datasets, and traces.

Listing Prompts

To see a formatted table of all available prompts:

coaia fuse prompts list

Getting a Specific Prompt

Retrieve a prompt by name. By default, it fetches the version with the latest label.

coaia fuse prompts get <prompt_name>

Options:

• --label <label>: Fetch the version with a specific label (e.g., dev, staging).
• --prod: A convenient shortcut for --label production.
• --json: Output the raw JSON response.
• -c, --content-only: Output only the raw prompt content, ideal for scripting.
• -e, --escaped: Output the prompt content as a single, JSON-escaped line. This is useful for embedding the content in other scripts or commands. Using -e implies -c.

Examples:

# Get the latest version of a prompt
coaia fuse prompts get MyPrompt

# Get the production version of a prompt
coaia fuse prompts get MyPrompt --prod

# Get only the content of a prompt
coaia fuse prompts get MyPrompt -c

# Get the content as an escaped, single line
coaia fuse prompts get MyPrompt -e

Managing Datasets

Listing Datasets

To see a formatted table of all available datasets:

coaia fuse datasets list

Getting a Specific Dataset and its Items

Retrieve a dataset's metadata and all of its items in a formatted display.

coaia fuse datasets get <dataset_name>

Options:

• --json: Output the raw JSON for the dataset and its items.
• -oft, --openai-ft: Format the dataset for OpenAI fine-tuning (JSONL).
• -gft, --gemini-ft: Format the dataset for Gemini fine-tuning (JSONL).
• --system-instruction "<text>": Customize the system instruction for fine-tuning formats. The default is "You are a helpful assistant".

Examples:

# Get a formatted view of a dataset and its items
coaia fuse datasets get MyDataset

# Get the raw JSON for a dataset
coaia fuse datasets get MyDataset --json

# Export a dataset for OpenAI fine-tuning
coaia fuse datasets get MyDataset -oft > training_data.jsonl

# Export for Gemini with a custom system instruction
coaia fuse datasets get MyDataset -gft --system-instruction "You are a creative writing assistant."

Creating a New Dataset

You can create a new, empty dataset directly from the CLI.

coaia fuse datasets create <new_dataset_name>

Adding Items to a Dataset

You can add new items (with an input and an optional expected output) to an existing dataset.

coaia fuse dataset-items create <dataset_name> --input "User question or prompt." --expected "Ideal model response."

Traces & Observations - Enhanced AI Pipeline Support

CoAiAPy provides comprehensive support for Langfuse traces and observations with enhanced pipeline integration.

Creating Traces

Create a new trace with session, user metadata, and optional environment variable export:

coaia fuse traces create <trace_id> -s <session_id> -u <user_id> -n "Trace Name"

Pipeline Integration Example:

# Create trace and export environment variables for pipeline use
eval $(coaia fuse traces create $(uuidgen) -s $(uuidgen) -u pipeline-user -n "AI Workflow" --export-env)
echo "Created trace: $COAIA_TRACE_ID"

Adding Observations

Add single observations to traces with auto-generated IDs and enhanced CLI options:

Basic Usage:

# Observation ID is auto-generated if not provided
coaia fuse traces add-observation <trace_id> -n "Processing Step" -i '{"input":"data"}' -o '{"result":"output"}'

# With explicit observation ID
coaia fuse traces add-observation <trace_id> <observation_id> -n "Custom Step"

Observation Types with Shortcuts:

# EVENT (default) - discrete events
coaia fuse traces add-observation <trace_id> -te -n "Data Loaded"

# SPAN - operations with duration  
coaia fuse traces add-observation <trace_id> -ts -n "Main Processing"

# GENERATION - AI model calls
coaia fuse traces add-observation <trace_id> -tg -n "LLM Response" --model "gpt-4"

Parent-Child Relationships:

# Create parent SPAN
eval $(coaia fuse traces add-observation $COAIA_TRACE_ID -ts -n "Main Workflow" --export-env)
parent_span=$COAIA_LAST_OBSERVATION_ID

# Add child observations under the SPAN
coaia fuse traces add-observation $COAIA_TRACE_ID -n "Step 1" --parent $parent_span
coaia fuse traces add-observation $COAIA_TRACE_ID -n "Step 2" --parent $parent_span

Pipeline Workflow Example:

#!/bin/bash
# Complete AI pipeline with automatic ID propagation

# Step 1: Create trace and export environment
eval $(coaia fuse traces create $(uuidgen) -s $(uuidgen) -u ai-pipeline --export-env)

# Step 2: Create main SPAN observation
eval $(coaia fuse traces add-observation $COAIA_TRACE_ID -ts -n "AI Processing Pipeline" --export-env)
main_span=$COAIA_LAST_OBSERVATION_ID

# Step 3: Add processing steps under the main SPAN
eval $(coaia fuse traces add-observation $COAIA_TRACE_ID -n "Data Loading" --parent $main_span --export-env)
eval $(coaia fuse traces add-observation $COAIA_TRACE_ID -tg -n "Model Inference" --parent $main_span --model "gpt-4" --export-env)
eval $(coaia fuse traces add-observation $COAIA_TRACE_ID -n "Results Processing" --parent $main_span --export-env)

echo "Pipeline complete! Trace: $COAIA_TRACE_ID"

Batch Observations

Add multiple observations from JSON or YAML files:

# From file
coaia fuse traces add-observations <trace_id> -f observations.json

# From stdin with YAML format
cat observations.yaml | coaia fuse traces add-observations <trace_id> --format yaml

# Dry run to preview what would be created
coaia fuse traces add-observations <trace_id> -f observations.json --dry-run

Example JSON format for batch observations:

[
  {
    "name": "Data Processing",
    "type": "SPAN",
    "input": {"dataset": "user_data.csv"},
    "output": {"processed_rows": 1000}
  },
  {
    "name": "Model Training", 
    "type": "GENERATION",
    "parent_observation_id": "previous-observation-id",
    "model": "gpt-4",
    "usage": {"tokens": 150, "cost": 0.003}
  }
]

Environment Variables for Pipelines

CoAiAPy exports standard environment variables for seamless pipeline integration:

• COAIA_TRACE_ID: Current trace identifier
• COAIA_SESSION_ID: Current session identifier
• COAIA_USER_ID: Current user identifier
• COAIA_LAST_OBSERVATION_ID: Most recently created observation ID
• COAIA_PARENT_OBSERVATION_ID: Parent observation ID (when using --parent)

Usage Pattern:

# Commands with --export-env output only shell export statements (no JSON)
eval $(coaia fuse traces create $(uuidgen) --export-env)
eval $(coaia fuse traces add-observation $COAIA_TRACE_ID -ts -n "Process" --export-env)

# Use the exported variables in subsequent steps
coaia fuse traces add-observation $COAIA_TRACE_ID -n "Child" --parent $COAIA_LAST_OBSERVATION_ID

Advanced Features

Datetime Format Support:

• ISO format: 2025-08-17T14:30:22Z
• TLID format: 250817143022 (yyMMddHHmmss)
• Short TLID: 2508171430 (yyMMddHHmm, seconds default to 00)

Usage Information:

coaia fuse traces add-observation <trace_id> -tg -n "LLM Call" \
  --model "gpt-4" \
  --usage '{"prompt_tokens": 100, "completion_tokens": 50, "total_cost": 0.0025}'

Metadata and Levels:

coaia fuse traces add-observation <trace_id> -n "Error Handling" \
  --level ERROR \
  --metadata '{"error_type": "timeout", "retry_count": 3}'

Contributing

Contributions are welcome! Please open an issue or submit a pull request for any improvements or bug fixes.

License

This project is licensed under the MIT License. See the LICENSE file for more details.