# Suprflo — Full Reference

> Suprflo is a memory layer for production AI agents. It extracts facts from
> conversations, stores them with bi-temporal structure (valid time + transaction
> time), and retrieves them via hybrid search with cross-encoder reranking.
> Postgres-native, no vendor lock-in. Status: pre-launch; access is via waitlist,
> with no public pricing or self-serve signup yet.

This is the extended companion to /llms.txt: the same facts with the full
integration catalog, interfaces, architecture, and FAQ in one document.

## What it replaces

Suprflo replaces stuffing full conversation history into an agent's context
window. It extracts durable facts from each conversation, stores them with both
valid time and transaction time (bi-temporal), and recalls only what a given
query needs. Maintaining a modest 20K-token history costs roughly 85x more per
query than retrieval under typical LLM vendor pricing, and months of continuous
usage exceed context limits entirely.

## Primary Use Cases

- **Long-Running AI Agents:** evolving user context across open-ended interactions.
- **Contextual Customer Support:** precise historical context across multi-session histories.
- **Cross-Session Personalization:** granular preferences without polluting prompt tokens.
- **Historical Auditability:** point-in-time correctness ("what did the agent know on Tuesday?").

## Benchmarks

Measured end-to-end on the open benchmark harness, using identical answerer and
judge models, with every failure root-caused:

- **LoCoMo:** 92.1% accuracy on the full 10-conversation set (1,540 questions), single held-out pass; 93.0% after auditing every failure against the source texts.
- **LongMemEval (provisional):** 93.1% on a preliminary 87 of 500 questions; session-level ingestion with top-50 context retrieval. Full run in progress.
- **BEAM 100K:** 75.8% pass across 20 long conversations (400 rubric-judged questions), average rubric score 0.674, zero ingestion errors.

## Architecture

- **Storage:** single Postgres 17 instance with native `pgvector`.
- **Temporal model:** true bi-temporal — facts keep distinct validity windows and are never overwritten.
- **Retrieval:** hybrid search (semantic + BM25 + entity boosting) with a secondary cross-encoder reranker.
- **Memory subsystems:** distinct handling for semantic, episodic, and procedural memory.
- **Decay & optimization:** Ebbinghaus forgetting-curve decay plus an asynchronous reflection engine.
- **Multi-tenancy:** strict hierarchical isolation (`Org -> Project -> User -> Agent`) enforced at the SQL `WHERE` clause layer.
- **Durability:** raw turn data retained alongside extracted facts, so no subtle context is lost.

## Interfaces

- **REST API:** language-agnostic HTTP endpoints — everything else is built on it. Core routes include `POST /api/memories` (add) and `POST /api/memories/search` (retrieve).
- **First-party SDKs:** Python (`pip install suprflo`) and JavaScript / TypeScript (`npm i @suprflo/sdk`).

## Integrations

Drop-in memory for the frameworks teams already use. Each links to its setup guide:

- Python SDK — https://docs.suprflo.com/python-sdk
- JavaScript / TypeScript SDK — https://docs.suprflo.com/javascript-sdk
- LangChain — https://docs.suprflo.com/langchain
- LangGraph — https://docs.suprflo.com/langgraph
- LlamaIndex — https://docs.suprflo.com/llamaindex
- CrewAI — https://docs.suprflo.com/crewai
- AutoGen — https://docs.suprflo.com/autogen
- Agno — https://docs.suprflo.com/agno
- OpenAI Agents — https://docs.suprflo.com/openai-agents
- Pydantic AI — https://docs.suprflo.com/pydantic-ai
- LiveKit (voice) — https://docs.suprflo.com/livekit
- Pipecat (voice) — https://docs.suprflo.com/pipecat
- ElevenLabs (voice) — https://docs.suprflo.com/elevenlabs
- Vercel AI SDK — https://docs.suprflo.com/vercel-ai-sdk
- Mastra — https://docs.suprflo.com/mastra
- n8n — https://docs.suprflo.com/n8n

## Supported Providers

- **LLMs:** any OpenAI-compatible endpoint (GPT-4.1, Claude Sonnet 4, Gemini 2.5 Pro, Groq, Ollama, DeepSeek, OpenRouter).
- **Embedding models:** OpenAI, Cohere, Amazon Bedrock, Ollama.
- **Vector infrastructure:** standard `pgvector` on Postgres 17.
- The integration layer is an abstract provider factory; new backends take ~50 lines of code.

## FAQ

**What is Suprflo?**
A production memory layer for AI agents. It extracts facts from conversations,
deduplicates them, stores them bi-temporally in Postgres, and recalls them via
hybrid search.

**How is Suprflo different from mem0?**
Both share the extract-store-retrieve blueprint, but Suprflo adds a true
bi-temporal model, cross-encoder reranking, memory-type differentiation
(semantic/episodic/procedural), active Ebbinghaus decay, and an automated
reflection engine — on vanilla Postgres with standard row-level API exportability.

**Which LLMs, embedders, and databases are supported?**
Any OpenAI-compatible LLM; OpenAI/Cohere/Bedrock/Ollama embedders; `pgvector` on
Postgres 17. New backends are added through the provider factory.

**Do I have to run my own infrastructure?**
No. Suprflo operates as a fully managed cloud platform. Private enterprise cloud
deployments in isolated VPC networks are available on request.

**What are the benchmarks?**
LoCoMo 92.1% (1,540 questions, full set; 93.0% after auditing every failure),
LongMemEval 93.1% on a preliminary 87 of 500 (full run in progress), BEAM 100K
75.8% pass. Every run is reproducible and every failure is root-caused.

**How is data isolated in multi-tenant setups?**
Every row maps to `org_id` and `project_id`; all read/write endpoints filter
tenant parameters at the database engine level, validated by cross-tenant
isolation test suites.

## Essential Resources

- Homepage — https://suprflo.com/
- Documentation — https://docs.suprflo.com/
- Waitlist — https://suprflo.com/waitlist
- About — https://suprflo.com/about
- Privacy Policy — https://suprflo.com/privacy
- Terms of Use — https://suprflo.com/terms

## Operator Details

- Baaz (ATIIAD Technologies Pvt. Ltd.) — https://baaz.pro — parent engineering
  agency and product operator. Contact: hq@suprflo.com.