agentscope-ai / agentscope-ai/QwenPaw

[Feature] Add optional PowerContext pluggable long-term memory backend

未关闭
#7,079 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看
主要语言
Python
星标
34.9k
派生
3.1k
平均合并
1 天 15 小时
30 天内合并 PR
225

描述

## Summary

Add an optional PowerContext-backed long-term memory backend for QwenPaw through the existing BaseMemoryManager and memory_registry extension points.

Image

Image

## Motivation

The default local memory backend is suitable for single-machine use, but deployments may need a shared, persistent, explicitly scoped memory service. PowerContext provides cross-session continuity, scope isolation, citation-aware retrieval, and a remote HTTP boundary without coupling QwenPaw to a database driver.

## Proposed scope

- Add PowerContextMemoryManager as a selectable peer of ReMeLightMemoryManager.
- Support automatic pre-turn retrieval and non-blocking post-turn task_state persistence.
- Expose memory_search and memory_remember only when PowerContext is selected.
- Preserve PowerContext citation metadata in search results.
- Add optional Console configuration for URL, token, scope, timeout, and auto-search settings.
- Register dynamic tools with the unified governance registry.
- Keep the default backend and Agent main flow unchanged.

## Non-goals

This proposal does not add Handoff, Experience, Skill, Candidate, Artifact, or database-driver integration to QwenPaw. PowerContext remains an external HTTP service.

## Validation

The prototype has 181 QwenPaw unit tests passing, Console TypeScript and locale checks passing, and has been exercised against a local PowerContext service for remember/search, citation preservation, cross-session recall, and scope isolation.

A PR with the implementation will reference this issue.

贡献指南

打开贡献指南

评估

这个 Issue 还没有评估数据。

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。