I added self-hostable memory backend support for google adk using openmemory #3388
Replies: 2 comments
|
is it working like InMemoryService by adk, i can see whole conversation being pushed to <PAST_CONVERSATION> tag just like In memory service. can we define scope like we do in vertex memory bank where agent decides which part of conversation is worth saving as memory. |
|
This is awesome. Self-hosted memory with multi-sector embeddings and decay is exactly what people need when they don't want to be tied to Vertex AI. One heads up: since this ended up living in adk-python-community rather than core the google.adk.memory import and pip install google-adk[openmemory] line aren't quite accurate anymore, might be worth a quick edit so folks don't hit an import error. @akshatladdha16 that scoping question is probably best asked over in the community repo where it can get into the internals. Closing this out as Show and Tell since it's a community integration rather than a core ADK issue, great first contribution keep it up. |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Hey everyone! Just finished my first major contribution pull request to Google's ADK and wanted to share.
What I built: Self-hosted memory backend support using OpenMemory - basically giving AI agents long-term memory without needing cloud services and it works exactly like Vertex AI memory bank.
ADK only supported Vertex AI memory before, which meant you needed Google Cloud to give your agents memory. Now you can run everything locally or on your own infrastructure.





Here's the usage - super simple:
Or just use the CLI:
adk web agents_dir --memory_service_uri="openmemory://localhost:3000"Cool features:
Install:
This is my first big open source contribution so any feedback would be awesome
All reactions