Skip to main content

Remembr

Remembr is persistent memory infrastructure for AI systems. It gives agents a clean store -> search -> delete loop, session-aware short-term context, scoped multi-tenant isolation, and a self-hosted stack that starts locally with one command.

Who it is for​

  • Application developers building assistants, copilots, and internal tools that need memory without inventing a memory layer from scratch.
  • Agent framework users working in LangChain, LangGraph, CrewAI, AutoGen, LlamaIndex, Pydantic AI, OpenAI Agents, or Haystack.
  • Platform teams self-hosting AI infrastructure with clear security, auditability, and deletion controls.

60-second demo​

cp .env.example .env
bash scripts/docker-init.sh
import asyncio

from remembr import RemembrClient, TagFilter


async def main() -> None:
    async with RemembrClient(
        api_key="rk_demo",
        base_url="http://localhost:8000/api/v1",
    ) as client:
        session = await client.create_session(metadata={"app": "demo"})

        episode = await client.store(
            "Ada prefers weekly billing summaries on Fridays.",
            role="user",
            session_id=session.session_id,
            tags=["kind:preference", "customer:ada"],
        )
        print(episode.embedding_status)

        results = await client.search(
            "When should billing summaries be sent?",
            session_id=session.session_id,
            tag_filters=[TagFilter(key="kind", value="preference")],
            search_mode="hybrid",
        )
        print(results.results[0].content)


asyncio.run(main())

Core ideas​

  • Sessions group a conversation or workflow run.
  • Episodes are immutable memory entries.
  • Search can be semantic, keyword, or hybrid.
  • Embeddings are asynchronous, so freshly stored episodes may return embedding_status="pending" before they become searchable semantically.
  • Deletes are soft by default so teams can restore by mistake window, audit, and purge later.

Start here​