fluxmem-sdk 0.2.0

Python SDK for FluxMem — dynamic memory + wiki layer for AI systems

fluxmem-sdk

Python client for FluxMem — dynamic memory + wiki layer for AI systems.

pip install fluxmem-sdk

Quickstart

from fluxmem_sdk import FluxMem

fm = FluxMem(api_key="fxm_your_key", namespace="user:default")

# Memory
fm.add(content="User prefers Python", memory_type="procedural")
results = fm.search("preferences")
context = fm.get_context("Tell me about the user", max_tokens=2000)

# Wiki / Articles
article = fm.articles.create(
    title="Deploying to GCP",
    body="## Steps\n\n1. ...\n2. ...",
    article_type="procedure",
    tags=["deploy", "gcp"],
)
articles = fm.articles.search(query="how do we deploy?")
fm.articles.feedback(article["article"]["id"], signal="helpful")

# Unified search across memories + articles
mixed = fm.unified_search("what does the user prefer?")

Async

from fluxmem_sdk import AsyncFluxMem

async with AsyncFluxMem(api_key="fxm_...", namespace="user:default") as fm:
    await fm.add(content="...", memory_type="semantic")
    results = await fm.search("...")
    await fm.articles.create(title="...", body="...", article_type="fact")

Features

• Memory layer: episodic / semantic / procedural / deductive with lifecycle (decay → archive → tombstone)
• Wiki layer: typed articles (fact / procedure / decision / playbook / anti-pattern / episode) with maturity transitions driven by feedback
• Hybrid retrieval: dense + sparse + graph + entity boost via RRF fusion
• Unified search: one query across memory + wiki corpora
• Implicit feedback: signed context_token per search hit — agents mark articles as used/unused to update ranking

Docs

• Quickstart: https://docs.fluxmem.org/quickstart
• API reference: https://docs.fluxmem.org/api
• Architecture: https://docs.fluxmem.org/architecture

License

MIT