Productivity

LILT

by LILT

Overview

Turn ChatGPT into a secure, natively multilingual workplace with LILT—enterprise infrastructure for verified translation, custom AI models, and expert human review. LILT connects your chat interface directly to your brand’s language assets, automated quality gates, and enterprise translation workflows via the Model Context Protocol (MCP). Generate verified, policy-compliant content, enforce corporate glossaries, and route high-stakes assets to professional human reviewers without ever leaving ChatGPT. Key Capabilities: - Real-Time AI Verification: Automatically validate translations for accuracy, terminology, and brand voice using built-in AI QA agents. - Calibrated Quality Gates: Run instant AI validation for daily communications, or trigger expert human review for mission-critical content. - Custom Enterprise Models: Translate using domain-adapted LLMs tuned specifically to your company terminology and style guides. - Live Terminology Sync: Access corporate glossaries and translation memories directly inside your chat thread. - In-Chat Project Management: Launch translation jobs, track pipeline progress, and pull completed assets straight into your conversation. - Enterprise Security & Governance: Enforce data isolation, role-based access control (RBAC), and full audit trails.

Tools

lilt_discover

ChatGPT

Discover available tools and bundles for the LILT API. - If no name or bundle_name is provided: Returns a list of available bundles as Markdown. - If a target matches a bundle (e.g. 'translate', 'analyze'): Returns the instructions and a lightweight index of available tools in that bundle with brief 1-line descriptions as Markdown. - If a target matches a tool (e.g. 'translate_list_files'): Returns the full Markdown schema, including detailed description, parameter definitions, and return values. Args: name: The name or list of names of bundles or specific tools to inspect. Leave empty to see all bundles. bundle_name: Keyword-argument fallback for backward compatibility. Returns: result (str): A Markdown string detailing bundles or tools.

lilt_execute

ChatGPT

Execute a specific tool by name. Args: tool_name: The exact name of the tool to execute. parameters: A JSON string or dictionary of arguments. Returns: result (str): The result of the tool execution as a string.

search_by_context

ChatGPT

Search for jobs, projects, documents, and linguistic entries across Lilt. Use this tool when you want to look up, locate, or query any entity or translation asset by partial/exact name or phrase matching. The tool searches across multiple indices including document filenames (e.g., searching for '*.json' returns all JSON documents, or '.pdf' for PDF files), document names, segment content, and linguistic assets. It queries Lilt's Context Database (bypassing slow MySQL for fast multi-index federated search) and returns matching Jobs, Projects, Documents, Segments (with hydrated termbase/resource info), and User Resources (translation memories and termbases) using high-performance cursor pagination. Args: query: The search term or name keyword to look up (e.g. "marketing"). If you are searching for any specific administrative or linguistic ID (e.g., specific Job ID, Project ID, Document ID, Memory ID, or Segment ID), you MUST pass it under the corresponding list parameter (job_ids, project_ids, document_ids, memory_ids, or segment_ids) rather than using the query string, as Elasticsearch global term scans do not index numeric entity identifiers as searchable text. first: Maximum number of search items to return per sub-category. Cap: 100. Default: 10. after: Base64-encoded cursor token to fetch the next page of results. Default: None. domain_id: Optional domain ID filter. creation_start: Optional ISO 8601 start date for creation filtering (e.g. '2026-01-01'). creation_end: Optional ISO 8601 end date for creation filtering. last_updated_start: Optional ISO 8601 start date for last updated filtering. last_updated_end: Optional ISO 8601 end date for last updated filtering. job_ids: Optional list of specific Job IDs to retrieve or filter by. project_ids: Optional list of specific Project IDs to retrieve or filter by. document_ids: Optional list of specific Document IDs to retrieve or filter by. memory_ids: Optional list of specific Translation Memory IDs to filter by. segment_ids: Optional list of specific Segment string IDs to retrieve. is_confirmed: Filter segments by confirmation status (True/False). is_reviewed: Filter segments by reviewed status (True/False). is_target_from_user: True to isolate human-edited/verified segments. translator_user_ids: Filter segments by translator/author IDs. assignee_emails: Filter documents by assigned collaborator email addresses. job_submitters: Filter jobs by submitter email addresses. source_languages: Filter results by source language using LILT internal codes (e.g. ['en']). CRITICAL: internal codes can differ from ISO standards (Chinese Traditional is 'zt', NOT 'zh-TW') — NEVER guess; call resolve_language_code first for every language the user names. A bare code automatically matches all its locale variants. target_languages: Filter results by target language using LILT internal codes (e.g. ['de', 'zt']). Same rule as source_languages: resolve names/ISO codes via resolve_language_code before searching. metadata_filters: Filter jobs or projects using custom key-value tag dictionaries, e.g. [{'key': 'tier', 'value': 'gold'}]. include_facets: Set to True to retrieve real-time analytics distribution facets. Returns: str: A pretty-printed markdown representation of the matching items across all indices, or an informative failure message.

util_upload_file

ChatGPT

Request an upload. Inline when content is set, otherwise mint both a presigned PUT URL (for callers that can run a shell command) AND a browser upload-page URL (for callers that prefer to drop files in a browser). The agent picks whichever path matches the user's context; the unused handle expires harmlessly. Args: filename: Filename with extension (e.g. "doc.pdf"). Required for every path — names the file server-side for inline uploads, and is needed to mint the presigned PUT URL. Inline mode splits on the final "."; names without a dot default to ".md". content_type: MIME type that the file will be uploaded with. CRITICAL: For CSV files (e.g., termbases, glossaries, or TMs), you MUST set this parameter to exactly "text/csv". Defaults to "text/plain". Ignored for non-CSV inline paths. content: UTF-8 text content for inline uploads. When set, no presigned URL or browser session is minted, and filename names the file server-side. Returns: dict: Shape depends on content. Inline mode (content is set) — terminal: - file_id (int | None): LILT file id of the inline upload. - file_metadata (dict): Full LILT file row. Dual-handle mode (content is not set) — followup required. If one path's mint fails the other is still returned; the failed path's fields are None. Presigned-URL path: - upload_id (int | None): Pass to util_wait_for_upload_completion(upload_id=...). - upload_url (str | None): One-shot presigned PUT URL. - method (str | None): HTTP method for upload_url. - content_type (str | None): Content-Type the URL was issued with. - curl_command (str | None): Ready-to-paste curl invocation; replace LOCAL_FILE. - file_location (str | None): Eventual storage location. Browser path: - session_id (str | None): Pass to util_wait_for_upload_completion(session_id=...). - browser_upload_url (str | None): URL to open in a browser for drag-and-drop upload. - expires_at (str | None): ISO-8601 timestamp after which the session is rejected. Always: - instructions (str): Guidance for picking a path.

util_wait_for_upload_completion

ChatGPT

Wait for an async upload to finalize. Pass exactly one of upload_id (user ran the presigned PUT) or session_id (user uploaded via the browser page). Routes to the matching poller and returns a unified shape. Args: upload_id: Identifier returned by util_upload_file alongside curl_command. Mutually exclusive with session_id. session_id: Identifier returned by util_upload_file alongside browser_upload_url. Mutually exclusive with upload_id. timeout_seconds: Maximum seconds to wait before returning status="TIMED_OUT". Defaults to 300. poll_interval_seconds: Delay between successive status polls. Defaults to 2.0 seconds. Returns: dict: Unified shape across both paths. - mode (str): "curl" or "link" — echoes the input id. - status (str): One of READY, FAILED, TIMED_OUT, NOT_FOUND. - file_ids (list[int]): IDs of finalized files. Curl path returns at most one; browser path returns all session files. - files (list[dict]): Full LILT file rows for the finalized uploads — id, name, category, labels, file_hash, detected_lang, timestamps. Aligned 1:1 with file_ids. Empty on non-READY outcomes, or if the metadata fetch on the curl path failed (the upload itself is still authoritative). - upload_id (int | None): Echoed when curl path, else None. - session_id (str | None): Echoed when browser path, else None. Raises: ValueError: If neither or both of upload_id and session_id are provided, or if a polling parameter is non-positive.