flumes-ai 0.2.3

Flumes – unified memory infrastructure SDK for AI agents (Python)

Flumes AI – Python SDK

Flumes AI is a unified memory infrastructure for LLM-powered agents and applications. This SDK lets any Python or LangChain stack plug into the Flumes Memory API in three lines of code.

👉 Learn more at https://www.flumes.ai/ – full docs at https://docs.flumes.ai/.

---

🚀 Installation

pip install flumes-ai  # includes the OpenAI client by default

Python ≥3.8. No system packages or C-extensions required.

---

🔑 Configuration

Environment variables (preferred):

Service	Variable	Example
Flumes	FLUMES_API_KEY	export FLUMES_API_KEY=sk_live_…
OpenAI	OPENAI_API_KEY	export OPENAI_API_KEY=sk-…

You can also pass api_key="…" directly to MemoryClient.

---

⚡ Quick-start (entity-first; no summarize route)

from flumes import MemoryClient

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

res = u.add("Planning a trip to France. Like wine, cheese and calm places.")
print(res.get("context", {}).get("summary"))

hits = u.search("trip recommendations", top_k=24)
print(len(hits.get("matches", [])))

Behind the scenes Flumes:

1. Adds the fact to long-term memory (summarised + deduplicated).
2. Retrieves relevant memories with semantic search.
3. Injects them into the LLM prompt (OpenAI by default).
4. Stores the assistant response for future context.

---

🛠️ Advanced (thin pass-through)

from flumes import MemoryClient

mc = MemoryClient(api_key="YOUR_KEY")
mc.search("wine", entity_id="user_42")
mc.get_all(entity_id="user_42", limit=50)

All endpoints map 1-to-1 with the REST reference.

---

✨ Key features

• Semantic search & vector similarity out-of-the-box.
• Automatic summarisation & deduplication (infer=True by default).
• Pluggable LLM backend – OpenAI today, Claude/Mistral coming.
• Structured logging – JSON events (memory.add.request, llm.called) for easy observability.
• Timeout & retry helpers – avoid first-call latency issues.

---

🗺 Roadmap

• Async transport with httpx.AsyncClient
• CLI: flumes chat, flumes memories list
• Policy plug-ins: custom summariser / retention strategies
• Local in-process transport (flumes-core) for offline testing

---

🤝 Contributing

1. Fork the repo and create a feature branch.
2. make install && make test
3. Open a PR – we ❤️ community help!

Please see LICENSE (MIT).

---

Made with 🧠 by the Flumes team – join us on the journey to give every agent a memory!