The three backends
Raw — on-pod memory files
Persistent Markdown memory stored on the pod’s own box. Private to the pod, nothing leaves it, and it’s destroyed with the pod. This is the default for Starter and for BYOK pods.
Lossless Context
Keeps the whole conversation in view instead of dropping older turns as a chat grows — so the agent never silently forgets earlier context in a long session. It works the running context harder (it uses your LLM), and can be layered on top of the other backends.
Semantic memory (cross-session)
The agent recalls facts across every conversation, not just the current one. Memories are isolated to this pod and stored privately. Powered by mem0, brokered so the key never touches the VM.
What you get by tier / mode
Advanced semantic recall is rolling out progressively. When the platform’s memory feature isn’t live yet, pods run on raw on-pod memory and the Memory tab notes that cross-session semantic memory and Lossless Context are coming soon.
BYOK memory options
When you deploy a BYOK pod, the Memory section lets you choose:- Raw files (default) — on-pod Markdown memory; nothing leaves the box.
- mem0 (your key) — cross-session semantic memory in your own mem0 account. Get a key at app.mem0.ai and paste it at deploy.
- Lossless Context — a checkbox to never lose earlier context in long chats (uses your LLM). Independent of the backend choice.
Viewing memory
The Memory tab shows the active backend (Persistent vs Semantic), whether Lossless Context is on, and a short explanation of what’s retained and where. If you’re on raw memory, it points to the upgrade path (deploy a managed Max pod, or bring a mem0 key on BYOK).Related
Deploy
Pick a tier / BYOK memory.
Manage
Where the Memory tab lives.
Models
The model that reads memory.