​

Flumes SDKs

​

What you get

• Entity‑first design: always scope by entity_id (user/session) and optional namespace.
• Add → Extract → Store in a single call using context assemble (no separate summarize step).
• Hybrid retrieval with presets (balanced, factual, recent, graphy) or explicit weights.
• Budget caps with pack telemetry on every add.
• Explicit LLM integration: the Agent helper requires an openai_api_key (or a custom backend).

​

Install

pip install flumes-ai           # core SDK
# optional Agent helper – requires OpenAI client
pip install "flumes-ai[agent]"   # or: pip install openai

​

Minimal example (Python)

from flumes import MemoryClient

client = MemoryClient(api_key="FLUMES_KEY", agent_id="travel-bot", namespace="prod")
u = client.for_entity("user_001")

res = u.add("Planning a trip to France. Like wine and quiet places.", budget="standard")
print("pack=", res.get("pack"))

hits = u.search("trip recommendations", top_k=8)
print("matches=", len(hits.get("matches", [])))

​

What happens on add()

• Flumes extracts structured facts/events from your turn.
• It stores them and returns pack telemetry (target/hard cap/used/dropped).
• You can later retrieve with search() or assemble full context via the API.

​

Optional Agent helper (explicit key)

from flumes import MemoryClient, Agent

client = MemoryClient(api_key="FLUMES_KEY", agent_id="travel-bot", namespace="prod")
agent = Agent(
    agent_id="travel-bot",
    entity_id="user_001",
    memory_client=client,
    openai_api_key="OPENAI_KEY",  # required explicitly
)

print(agent.chat("Plan my trip."))

​

Next: Python details

• Dive into the Python specifics, including retrieval knobs and budgeting, in the Python tab.