Quick Start

Build your first agent: identity → private memory → shared domain → recall.

1. Install & set a key

pip install "unibase-membase-sdk[recovery,runtime] @ git+https://github.com/unibaseio/unibase-membase.git"
export MEMBASE_PRIVATE_KEY="0x..."     # or omit and use Wallet.generate()

The [recovery,runtime] extras enable memory.ingest / memory.answer. Drop them for raw KV + cooperation only.

2. Create an agent

from unibase_membase import Membase, Wallet

agent = Membase(Wallet.from_env())     # identity = wallet address
print(agent.address)

3. Private memory

Only this wallet can decrypt — a single-member domain.

agent.private().set("profile/lang", {"value": "zh"})
print(agent.private().get("profile/lang"))   # → {"value": "zh"}

4. Shared memory across agents

alice = Membase(Wallet.from_env())
bob   = Membase(Wallet.generate())

team = alice.create_domain("team:engineering")
team.set("roadmap/q3", {"goal": "ship the agent-direct evolution"})

code, _ = alice.invite_to(team)              # ECIES-wrapped key
bob_team, _ = bob.accept_invite(code)
print(bob_team.get("roadmap/q3", author=alice.address))
# → {'goal': 'ship the agent-direct evolution'}

5. Recall (ask about the past)

agent.memory.ingest(turns=[
    {"role": "user", "content": "I'll go to Tokyo in November 2026."},
    {"role": "user", "content": "Actually, moving the trip to August 2026."},
])
print(agent.memory.answer("When is the Tokyo trip?", query_date="2026-08-01"))
# → {'answer': 'August 2026', ...}

Next

• Memory & Recall · Domains & Encryption
• Cooperation Protocol — multi-agent sessions & games
• Settlement — on-chain metering
• Runnable examples: examples/