xAgent 0.0.6.beta Is Available: More Reliable File Handling and Long-running Tasks

xAgent 0.0.6.beta is now available. This release focuses on a foundational part of task quality: making the path from file upload to reading, processing, and delivery more reliable.

xAgent is not designed to simply attach files to a chat. It is designed to let Agents keep working on the server. For meeting notes, contracts, reports, spreadsheets, presentations, and operational materials, correct recognition, location, and reuse of files directly affect whether a task can be completed reliably.

This is still a beta release. This post explains what is changing for users preparing to deploy or upgrade; it does not replace the formal installation and upgrade instructions.

Files and Attachments: Validate Before a Task Uses Them

0.0.6.beta unifies how session attachments and workspace files enter the system. A file is accepted only after format, size, and content parsing checks pass, so a failed upload does not leave behind a record that looks available but cannot be used.

The current supported formats are DOCX, text-extractable PDF, PPTX, XLSX, CSV, TSV, static HTML, Markdown, common text and code files, and PNG, JPEG, and WebP images. Upload limits are now handled by file type and entry point, with clear errors before a file is saved instead of one universal size limit.

For long documents, xAgent no longer puts the whole file into the context at once. Small text files can be read in full; larger files expose structured locations so an Agent can read the relevant pages, sheets, slides, or sections when needed. This preserves context and makes it easier for the model to return to the right source material.

Word and HTML files preserve headings, paragraphs, emphasis, lists, links, and tables where possible. PDFs, spreadsheets, and presentations retain page, worksheet, and slide locations. This release also fixes workspace files that appeared to exist but could not be read, as well as download filename issues involving Chinese characters or spaces.

GIF files, legacy Office formats, encrypted documents, and scanned PDFs without a text layer are not supported for upload yet.

Word Documents: From Completed Content to Delivery Validation

This release adds a built-in Word template library and a default template. Documents can use either built-in or user-provided templates while retaining template styles, headers, footers, and placeholders.

The generation flow now favors completing the outline and full content before creating the document, reducing layout drift from repeated edits. Before delivery, xAgent can validate a DOCX by rendering it to PDF and page images, checking pagination, fonts, spacing, tables, headers, footers, and text clipping.

xAgent also adds shared font resources. The first bundled fonts are Noto Sans SC, Noto Serif SC, and Noto Sans Mono CJK SC. Administrators can add local fonts or register internal online fonts with size and integrity checks.

Sub-sessions: Fewer Steps Before Collaboration Starts

When creating a sub-session, xAgent now performs initial orchestration automatically and selects an appropriate Agent and available capabilities for the task. This setup no longer requires a separate approval, reducing the wait before collaboration begins.

The release also improves capability merging and deduplication to avoid loading the same Tool repeatedly. Sub-sessions can recognize the active collaboration request and reply to it, while already-completed delegated work is not rerun or sent again during follow-up handling.

Models and Runtime: Fewer Hidden Configuration and Execution Problems

Model request caching is more stable in ongoing conversations, reducing unnecessary cache invalidation. Configuration compatibility for OpenAI, Anthropic, and Gemini is improved, and changes take effect sooner after adding, editing, changing the default model, or deleting a model configuration.

On the CLI side, Python dependencies are installed in a user-level shared location so sessions for the same user can reuse them while built-in runtime resources remain read-only. Relative paths resolve from the command working directory; use $XAGENT_WORKSPACE when referencing the workspace root.

Before execution output reaches the Agent, history, or UI, host-machine and temporary paths are hidden in favor of stable logical paths. The CLI also enables pipeline failure detection by default and improves failure analysis, useful retries, and result validation.

Updates and Data Cleanup

For supported managed installations, the console can start an update after detecting a new version. The system downloads, switches, and restarts automatically, and rolls back if the health check fails. Older OpenAI model configurations are adapted to the current type during upgrade while retaining existing connection parameters and model settings where possible.

Historical file migration is also more resilient: one damaged or unrecognized legacy file does not block service startup. Formatting user data now more completely removes triggers, sessions, attachments, workspaces, and related runtime data. Existing login sessions are invalidated and a new main session is created after the user signs in again. Accounts, model usage, and explicitly retained external connection information are not affected.

Preparing to Upgrade

Back up the runtime directory, configuration, database, and Connector state before upgrading. See the Changelog and Start Installation for installation and upgrade steps.

Report deployment, file handling, model configuration, or Connector integration issues through Issues.