Durable memory
Per-user memory that survives sessions, deploys and model swaps.
Every turn is stored in a per-tenant vector memory, searchable by meaning. Seed it from your other systems, search it from your own code and erase it when a user asks to be forgotten.
Memory · user_42
- strong
A walk after work helps them reset on bad days
memory_type: fact
- moderate
Prefers short replies late at night
memory_type: preference
- weak
Sister visiting next weekend
memory_type: note
Per-tenant vector store · removed by erase_user_data()
Memory the model can use and you can inspect
Humane stores each interaction in the user's memory and pulls the most relevant items when it builds context for a reply. Search results come back with a match strength and the reason they matched, so you can see why the model was reminded of something.
- Semantic search per user, or most recent first when the query is empty
- Seed facts and notes from intake forms, CRMs or an older system
- Manual inserts don't inflate the familiarity signal
- Stored on a persistent volume, so memory survives deploys
Context · user_42
- system_personakept
- behavioral_constraintskept
- relational_context
- recent_conversation
- memories1 dropped over budget
- active_tasks
- time_situation
- response_instructionskept
tone
warm_supportive
modifier
cautious (grudge)
empathy
0.82
formality
0.35
response_length
short
Seed and search
client.create_memory(
user_id="user_42",
text="Prefers short replies late at night. A walk after work helps.",
memory_type="preference",
metadata={"source": "onboarding_form"},
)
hits = client.search_memory("user_42", query="what helps when stressed", limit=5)
for m in hits["memories"]:
print(m["match_strength"], m["text"])Built to be forgotten on request
- erase_user_data() removes database rows and the user's vector memory
- A failed vector delete is reported as an error, not a silent success
- export_user_data() returns what Humane holds on a user
- Metadata is cleaned before it reaches the vector store
Memory questions
Something else? Ask us.
Works well with
Journey and history
The journey endpoint returns everything that concerned one user in time order, from their messages to safety events, proactive triggers and experiment assignments. Conversation history can be read or cleared per user.
ExploreTone and context
Before your model runs, Humane assembles its prompt from the user's state, relationship, recent conversation, relevant memories and open tasks, scoring and trimming each section to fit a budget. It also sets the tone, empathy, formality and length the reply should take.
ExploreUser state across sessions
Humane keeps a running state for every end user, from mood and energy to trust and grudge, each with its own behaviour over time. Your model finally knows whether it is talking to someone on a good day or at the end of a bad week.
ExploreEvery AI should know who it's talking to.
The safety and human-oversight layer for emotional AI: user state across sessions, risk alerts, safety gates and human takeover, with every decision explained.
