Token-budgeted conversational memory for Spring AI. The library stores accumulated learnings on the filesystem, injects a budgeted subset into each ChatClient request, and optionally summarizes older entries with a cheaper model when the uncompacted set crosses a configurable threshold.

See the Agent Memory documentation for the roadmap, module notes, and the originating research.

Current artifacts are 0.5.0 (memory-core and memory-advisor). Requires Java 17+ and Spring AI 2.0.1. This is a pre-1.0 library.

<dependency>
    <groupId>io.github.markpollack</groupId>
    <artifactId>memory-advisor</artifactId>
    <version>0.5.0</version>
</dependency>

The 0.4.0 filesystem store is local, plaintext, and single-writer.

• Memory is written to the local filesystem in clear text. It is not encrypted and is not shared storage.
• There is no locking, no atomic index replacement, and no crash recovery. A second concurrent writer, or a crash during an index write, can corrupt or truncate _index.json.
• Stored memory is injected verbatim into the model prompt, so write only trusted content to it.
• Multi-process or concurrent-writer use requires external coordination.

./mvnw clean verify

Standalone consumer Jackson resolution (no AgentWorks BOM):

./scripts/check-consumer-resolution.sh

Licensed under the Business Source License 1.1. Versions 0.3.0 and earlier remain available under the historical Apache License 2.0.