MateClaw v1.5.0 delivers engineering-ready workflows for open-source AI agent runtime

Recently, the open-source AI Agent Runtime framework MateClaw released version 1.5.0. Rather than simply expanding the number of integrated models, this update focused on strengthening the essential infrastructure needed for Agents to work effectively in real-world teams. Its core improvements centered on three key areas: goal acceptability, knowledge base consistency, and multi-user memory isolation.

In real business settings, traditional agents often relied on a vague "completion score" to gauge task progress, leaving managers unable to pinpoint where a task was blocked. To solve this opacity, MateClaw v1.5.0 introduced a "Goal Checklist" mechanism. This mechanism enables the system or running large models to dynamically break down a macro goal into multiple independent verification criteria. A built-in evaluator then checks each criterion at various stages and logs corresponding evidence. Only when every item on the checklist is passed does the system mark the task as complete. This digital checklist makes task boundaries clearer while providing precise contextual input for automated agent follow-up.

As the most significant engineering effort in this release, MateClaw evolved the traditional knowledge base (LLM Wiki) into a self-maintaining "Knowledge Engine." It now supports wiki-link functionality akin to Wikipedia, including cascading page renames, deletion cleanup, and broken link scanning to preserve consistency. More notably, the new system introduces a "knowledge hierarchy" that splits content into basic fact layers and experience summary layers. When a base "fact page" changes, all dependent "experience pages" are automatically flagged for review, addressing the classic problem of "facts changed, conclusions outdated." Additionally, the PageType Profile feature lets administrators configure structured fields and Markdown templates for each page type, and enforce fail-safe, fine-grained access control based on an "employee + knowledge base + page type" matrix.

To help AI agents integrate safely into multi-user collaboration environments, MateClaw v1.5.0 fully activated the multi-user memory isolation (Memory per-owner) feature. At its core, the system uses owner identifiers (owner_key) and three visibility scopes to ensure that the same agent serving multiple users through a web console, instant messaging (IM) channels, or third-party APIs never mixes personal privacy data or long-term memory.

In addition to these three main features, the new version delivers a series of production stability improvements. These include allowing employees to bind to a default Wiki knowledge base, optimizing the model selection chain so that preference provider routing works as intended, and adding a new Claude Opus4.8 model entry. On the engineering side, multimedia files generated by tools are now stored securely and automatically cleaned up via scheduled tasks. The MCP tool's default read timeout has been extended to 60 seconds to reduce false positives. Meanwhile, WeChat, Enterprise WeChat, and Feishu channels now feature a unified inbound media pipeline that determines file types using feature codes and implements exponential backoff retries, significantly improving the reliability of complex file interactions in group chats. The upgrade configuration of this version is fully compatible with historical data, and migrations are automatically executed by Flyway, accelerating AI agents' transition toward true industrial-level deployment.