multiai Release Notes

The release history is listed from newest to oldest. To check your installed version, run: python -m pip show multiai. To upgrade to the latest version, run: python -m pip install --upgrade multiai.

1.6.0 - 2026/6/16

• New: Added response attachment support. When a provider API returns actual file-like data in an AI response, multiai can now receive it, expose it through the Python API, and save it from the CLI.
  • Added the ResponseAttachment data structure.
  • Added client.response_attachments to access attachments returned by the latest response.
  • Added client.save_attachment(attachment, filename) to explicitly save a response attachment when using multiai as a Python library.
  • Added automatic saving of actual response attachments in the CLI.
  • Added [response_attachment] directory setting to configure the save directory. If not set, ./multiai_attachments is used.
  • Added --response-attachment-dir DIR CLI option to override the response attachment save directory.
  • If a response attachment is returned as a URL, multiai displays the URL but does not download it.
• Note: In the current chat-based API usage, most providers usually return text only, so response attachments are uncommon. This feature is mainly a foundation for future support of APIs and tools that can return actual files, such as image generation, code interpreter outputs, Gemini inline data, or OpenAI Responses API file outputs.

1.5.2 - 2026/6/10

• Added support for Anthropic’s Claude Fable model. The temperature parameter is now only available for Sonnet and Haiku models.

1.5.1 - 2026/5/1

• Not using the temperature parameter for Anthropic’s Claude Opus model, due to a specification change introduced in Claude Opus 4.7.

1.5.0 - 2026/1/19

• New: Added --list CLI option to list available models
• New: Added support for loading the Azure Speech API key for use with multiai-tts

1.4.3 - 2026/1/12

• New: Released multiai-tts, an official extension library adding Text-to-Speech capabilities via OpenAI and Google GenAI.

1.4.2 - 2025/12/24

• Replaced deprecated PyPDF2 with pypdf.

1.4.1 - 2025/12/18

• Migrated Gemini client from google.generativeai to google.genai as google.generativeai is being deprecated.

1.4.0 - 2025/10/10

• New: file attachments for text-like inputs (CLI and library). Special handling for txt/md/pdf/docx/html/htm/csv; unknown extensions accepted if valid UTF-8 text (no NUL bytes), otherwise rejected.
• New: CLI options -f/–file and –attach-limit (auto-summarizes oversized attachments with a notice); when no prompt is provided, attachments are applied only to the first turn before entering interactive mode.
• New: library helpers retrieve_from_file(), summarize_text(), and attach_char_limit config.
• Change: removed -f/–factual option (breaking change); added dependency on python-docx for .docx support.

1.3.2 - 2025/9/30

• Migrated from pkg_resources to importlib.metadata as pkg_resources is being deprecated.

1.3.1 - 2025/8/8

• Adapt to gpt-5 series

1.3.0 - 2025/5/2

• Resolve conflict with the -l option

1.2.0 - 2025/2/21

• xAI support

1.1.0 - 2025/2/6

• DeepSeek and local LLM support

1.0.0 - 2024/12/6

• First stable version

0.1 - 2024/8/16

• First release

See also

• Release at PyPI
• GitHub commits
• X post