Logs the user into the Adeu Cloud backend. Securely opens a browser window for authentication.

Logs out of the Adeu Cloud backend by clearing the local API key from the OS Keychain.

Accepts all tracked changes and removes all comments in a single operation, producing a finalized clean document. Use this when a document review is entirely complete and you want to clear all redlines. For selective acceptance/rejection of specific changes, use process_document_batch instead.

Compares two DOCX files and generates a text-based Unified Diff. Use this to see exactly what changed between two versions of a document. By default (compare_clean=True), it compares the 'Accepted' finalized states of both documents. Set compare_clean=False if you need to compare the raw underlying text including Tracked Change CriticMarkup.

Opens a local file in its native desktop application (e.g., Microsoft Word for DOCX files).

Applies a batch of structural edits, text modifications, and review actions to a document. This is your primary tool for editing DOCX files.

CRITICAL: All changes in the batch evaluate against the ORIGINAL document state. Do not send sequential edits that depend on each other within the same batch (e.g. rename X to Y, then modify Y). Instead, apply the rename in one batch, then modify Y in a subsequent batch.

The changes parameter is a list of operations. Each item MUST have a type:

1. 'modify': Search-and-replace text. Provide exact target_text (CRITICAL: include surrounding context if the word appears multiple times to ensure unique matching) and new_text (the replacement). new_text supports full Markdown structure: '# Heading 1' through '###### Heading 6' at the start of a line for heading styles, 'bold' and 'italic' inline formatting, and blank lines ('\n\n') to split new_text into multiple paragraphs. Multi-paragraph inserts are tracked as one logical revision. To delete text, make new_text empty. Do NOT manually write CriticMarkup tags ({++, {--, {>>). To add a comment, use the 'comment' parameter.

2. 'accept': Finalize a tracked change. Requires target_id (e.g., 'Chg:12'). (Note: Accepting one half of a paired modify cascades to accept the other half).

3. 'reject': Revert a tracked change. Requires target_id (e.g., 'Chg:12'). (Note: Rejecting one half cascades to reject the other half).

4. 'reply': Reply to a comment. Requires target_id (e.g., 'Com:5') and text.

5. 'insert_row': Insert table row. Requires target_text (anchor), position ('above'/'below'), and cells (Markdown strings).

6. 'delete_row': Delete table row. Requires target_text inside the row to be deleted.

Always provide a realistic author_name for Tracked Changes. This name will be used for attribution in the document's tracked changes and comments.

Reads a DOCX file and extracts its text content. Use this to ingest documents into your context window. By default (clean_view=False), it returns text with inline CriticMarkup (e.g., {++inserted++}, {--deleted--}, {==highlighted==}{>>comment<<}) representing Tracked Changes and Comments. Set clean_view=True ONLY if you want to read the final, clean text, ignoring all redlines and comments.

PAGINATION & OUTLINE:

• mode='outline' returns a structural map of headings with page numbers, styles, table presence, and referenced footnotes. Body content is omitted. Use this first on large documents to plan targeted reads.

• mode='full' (default) returns the document body. Documents over ~19,000 characters are split into pages; use page=N to read a specific page (1-indexed). Documents under the limit are returned in full on page 1.

• Page boundaries differ between clean_view=True and clean_view=False.

• The Structural Appendix (defined terms, anchors, diagnostics) is repeated on every page.

Creates an email draft in the user's native draft box (e.g., Outlook/Gmail). Can either start a NEW email, or REPLY to an existing thread. To REPLY, provide 'reply_to_email_id' (the short ID from search_and_fetch_emails). To start a NEW email, omit the ID but provide 'subject' and 'to_recipients'. Allows attaching local files (PDF/DOCX) by providing their absolute paths. The body should be formatted in Markdown.

Searches the user's live email inbox. By default, searches only the Inbox folder (matching what the user sees in their mail client) — this excludes deleted items, drafts, and spam. Use filters to find specific emails (e.g., 'is_unread=True' for new emails, 'days_ago=7' for last week, 'folder=sent' for sent items, 'folder=all' to search the entire mailbox including trash). It returns a list of lightweight email previews. To read the full email body, thread history, and automatically download attachments to local disk, call this tool again and provide the specific email_id. Emails often contain attachments. It is highly recommended to always provide the working_directory parameter so attachments are saved directly to the user's actual project folder. This directory path refers to the user's native operating system, not the LLM's sandbox environment.

Sanitizes a DOCX file by stripping dangerous metadata (rsids, author names, template paths, DMS metadata, hidden text, orphaned content) and producing an audit report of everything removed. Use this before sending documents to external parties. Supports three modes: full scrub (for signing/closing), keep-markup (preserves your track changes and open comments), or baseline (recomputes your delta against the original document).

Validates documents for inconsistencies, contradictions, and risk assessments. To START a new validation, provide 'file_paths' as a JSON-encoded string representing a list of file paths. This will immediately return a task_id. To CHECK the status of a validation, call this tool AGAIN and provide ONLY the 'task_id'. The checking process will poll for up to 50 seconds. If it times out, continue checking.