v0.6.2

this update includes a fix to avoid having the model to reread a file after it has edited it itself.

v0.6.1

0.6.0 introduced an auto mode.
this patch changes the interruption key from ESC to SPACE, because clack/prompts uses ESC already for canceling the CLI.

v0.6.0

Auto Mode

This release introduces an auto mode.
By launching md-ai with an --auto, your approval for write operations and executing commands becomes optional.
md-ai will wait --auto-timeout seconds (defaults to five) before it continues.
During this timeout you can press enter to proceed immediately or esc to have the same deny / approval interface as before.

You could now just let it run and do it's thing when you give it a task.

Provide reasons when denying operations

You are now asked to say why you choose to deny a request.
The model can use this information to take action immediately.

Other changes

The write files tool does now only accept a single patch operation instead of many.
The write files tool only allows writes on files the model has recently opened. If a file has changed since it was last opened (e.g. you edited it), the tool instructs the model to re-read the file.

Smaller fixes.

v0.5.0

Fetch URLs

This release adds a new tool for fetching URLs.
You can now paste documentation, blog article or other URLs into the chat and the LLM can use the tool to get the content.
It uses Mozillas Readability package to extract the main content from the fetch response.
The LLM can ask the tool to extract links from the page via regex, which will allow the LLM to "crawl" a site.

Breaking Changes

Library mode usage has changed. See the README for an up to date reference.

v0.4.0

Execute Shell Commands

md-ai now includes an exec-command tool. The model calls the tool with the command and an explanation. You can choose to allow or deny the execution or always allow it until the md-ai session is closed.

Open the Editor when it's the AI's turn

When it's the AI's turn you now can also choose to open the editor.

Fixes and Other Changes

• .gitignore parsing has been improved
• the default value from the grep search tool was removed. OpenAI was complaining about default values in tool schemas.
• the README now contains a usage guide for VS Code thanks to @MatthiasLettl

v0.3.0

• support for config file
• compress tool call and tool result payloads by default
• improve write file: receive JSON from model, ignore whitespaces, offer replace-file option, apply search and replace to entire file
• support system messages in markdown
• add a user message without opening a file

v0.2.0

Features

• the write files tool allows moving files
• a new search tool uses grep to let the ai search in files

Other

• add file (within writeFiles) fails if it would override an existing file
• src is published to npm