Suprflo — Integrations & SDKs

Suprflo is a bi-temporal memory layer for AI agents: it extracts facts from conversations, stores them with validity windows, and recalls exactly what your agent needs. Every integration below talks to the same REST API.

Get an API key

1. Sign in at suprflo.com.
2. Go to Dashboard → API keys → Create key.
3. Copy the key (shown once). It carries your organization + project scope.

Base URL & auth

• Base URL: https://api.suprflo.com
• Auth: send the key as a bearer token on every request:

  Authorization: Bearer <YOUR_API_KEY>

• All request/response bodies are JSON.

Core concepts

• Memory — a single extracted fact (e.g. "prefers window seats"), not the raw text.
• Subjects — every memory is scoped to a user_id (and optionally agent_id, run_id, app_id). Pass the same ids on add and search/get to isolate context.
• Add → extract → store: you send messages; Suprflo extracts the durable facts.
• Search: semantic + keyword retrieval returns the most relevant memories for a query.
• Bi-temporal: facts carry validity times and are never silently overwritten — the history endpoint shows every change.

Start here

Integration	Install	Docs
Python SDK	pip install suprflo	python-sdk.md
JavaScript/TypeScript SDK	npm install @suprflo/sdk	javascript-sdk.md
REST API	— (any HTTP client)	rest-api.md

Framework adapters (LangChain, LangGraph, LlamaIndex, CrewAI, AutoGen, Agno, OpenAI Agents, Pydantic AI, LiveKit, Pipecat, ElevenLabs, Vercel AI SDK, Mastra, n8n) each have their own guide — see the sidebar for the full catalog.

Self-hosting (advanced)

The hosted service at api.suprflo.com is fully managed — the sections above are all you need to build on Suprflo. This section is only relevant if you run the Suprflo backend yourself: it's configurable by environment variable, so you can swap the model or database it runs on without touching your application.

Component	Env var	Options
LLM (fact extraction)	LLM_PROVIDER	openai, anthropic, bedrock, groq, together, deepseek, xai, mistral, minimax, lmstudio, litellm, vllm, sarvam, gemini, azure_openai, ollama, langchain
Embedder	EMBEDDING_PROVIDER	openai, ollama, bedrock, azure_openai, gemini, vertexai, huggingface, together, lmstudio, langchain, fastembed
Reranker	RERANKER_PROVIDER	none, cross_encoder, cohere, huggingface, llm, zero_entropy
Vector store	VECTOR_STORE_PROVIDER	pgvector, qdrant, chroma, pinecone, weaviate, milvus, redis, mongodb, elasticsearch

Each provider reads its own credentials from its conventional env vars (e.g. OPENAI_API_KEY, QDRANT_URL). pgvector is the reference vector store and the only one that natively covers the whole engine surface — the others are functionally partial, and each store's class docstring states exactly how. Only pgvector is exercised against a real server today; the rest are unit-tested against mocked clients, so smoke-test one before enabling it in production.

Every provider has its own page with a config example and caveats. Browse them under Self-hosting in the sidebar: LLMs, Embedders, Rerankers, Vector stores.