v0.6.0 Announcement

English

Agent Memory Bridge v0.6.0 is out.

Two-channel MCP memory for coding agents:

• durable knowledge in memory
• coordination events in signal

This release pushes the bridge forward on learning quality without changing the MCP surface.

1. Reflex enrichment is now safer and smarter

• optional classifier-assisted enrichment
• shadow and assist rollout modes
• keyword/rule fallback stays in place
• fixture-based regression coverage for classifier-vs-fallback behavior

1. Retrieval remains measurable

• canonical benchmark
• precision@k
• expected_top1_accuracy
• current canonical fixture: memory_expected_top1_accuracy = 1.0

1. Signal lifecycle stays production-shaped

• claim -> extend -> ack / expire / reclaim
• extend_signal_lease remains part of the public MCP surface
• hard expiry stays a real upper bound

Also included:

• 73 passed
• terminal demo GIF in the repo

GitHub:
https://github.com/zzhang82/Agent-Memory-Bridge

Release:
https://github.com/zzhang82/Agent-Memory-Bridge/releases/tag/v0.6.0

Demo:
https://github.com/zzhang82/Agent-Memory-Bridge/blob/main/examples/demo/terminal-demo.gif

简体中文

Agent Memory Bridge v0.6.0 发布了。

这是一个面向编码代理的双通道 MCP 记忆层：

• memory 负责持久知识
• signal 负责协调事件

这次发布把重点放在 learning quality 上，而且没有改 MCP surface。

1. Reflex enrichment 更聪明，也更稳了

• 新增可选的 classifier-assisted enrichment
• 支持 shadow 和 assist rollout
• keyword / rule fallback 仍然保留
• 现在有 classifier-vs-fallback 的 fixture 回归测试

1. Retrieval 仍然是可度量的

• 有 canonical benchmark
• 会看 precision@k
• 也会看 expected_top1_accuracy
• 当前 canonical fixture 上：memory_expected_top1_accuracy = 1.0

1. Signal lifecycle 继续保持 production-shaped

• claim -> extend -> ack / expire / reclaim
• extend_signal_lease 仍然属于公开 MCP surface
• hard expiry 依然是严格上限

另外这版还包括：

• 73 passed
• repo 里的终端 demo GIF

GitHub:
https://github.com/zzhang82/Agent-Memory-Bridge

Release:
https://github.com/zzhang82/Agent-Memory-Bridge/releases/tag/v0.6.0

Demo:
https://github.com/zzhang82/Agent-Memory-Bridge/blob/main/examples/demo/terminal-demo.gif