F.·chat · live agent · gpt-4o-mini · via openrouter
Talk to an agent with persistent memory.
Each message you send is routed through Engram's pipeline. The backend retrieves the most relevant engrams from your stream, feeds them to the model, and writes whatever the model finds worth remembering back into the chain. The right-hand pane is your live memory.
// This agent has no memory yet. Tell it something it should remember.
// agent_id = …
// memory.size = 0
// pipeline · stream > embed > cosine top-k > llm > persist
// agent_id = …
// memory.size = 0
// pipeline · stream > embed > cosine top-k > llm > persist
head · 0x0…
// empty stream
// first engram lands here as soon as the agent learns something
// first engram lands here as soon as the agent learns something
how this works
Every message you send is routed through the Engram pipeline. The backend picks top-k relevant engrams from your stream and feeds them to the model. Whatever the model decides is worth remembering becomes a new engram on the chain.
live