Use Case

    AI Agent Memory with Endee

    Give autonomous agents a persistent long-term memory. Store experiences as vectors and retrieve the most relevant ones across sessions and across agents.

    CrewAILangChain AgentsLlamaIndexCross-session MemoryMulti-agent Shared MemorySemantic Retrieval

    Capabilities

    Built for stateful agentic AI

    Long-term Semantic Memory

    Agents store experiences, facts, and task outputs as vectors. On the next run they retrieve the most relevant memories semantically rather than by keyword or recency. Memory retrieval quality improves as the agent accumulates more experience over time.

    Persistent Across Sessions

    Context survives LLM context window limits and session restarts. Agents pick up where they left off using memory retrieved from Endee. This makes agents genuinely stateful across days, weeks, and months of operation rather than starting fresh each run.

    Native Framework Integrations

    Official integrations for CrewAI, LangChain, and LlamaIndex. Drop Endee in as the vector store with no architecture changes. The EndeeStorage adapter for CrewAI and the EndeeVectorStore for LangChain are available on PyPI with full documentation.

    Shared Team Memory

    Multiple agents in a crew share one Endee index. Tag memories by agent_id or memory_type to let agents collaborate with shared and private memory lanes. A research agent stores findings that a writing agent retrieves in the same pipeline run.

    Memory with Context Metadata

    Store rich metadata alongside each memory vector: agent_id, session_id, timestamp, memory_type, task_id, and confidence score. Filter retrieved memories by any combination of these fields so agents get contextually appropriate memories, not just semantically similar ones.

    Multi-modal Agent Memory

    Store not just text memories but also image analysis results, tool call outputs, and structured data as vectors. Agents that process images, run code, or call APIs can store and retrieve those experiences the same way as text-based memories.

    Process

    How it works

    1

    Store task output as a vector

    After each task or reasoning step, encode the output as a dense vector and write it to Endee with metadata including agent_id, session_id, timestamp, and memory_type. This persists the agent's work product regardless of context window size.

    2

    Retrieve relevant memories

    At the start of each run, embed the current task description and query Endee for the k most semantically similar past memories. Filter by agent_id to scope memory to the current agent, or query shared memory from peer agents.

    3

    Pass memories as context to the LLM

    Prepend the retrieved memories to the agent's system prompt or task context. The agent reasons over both its current task and its accumulated knowledge, producing better results without needing the full history in its context window.

    Integrations

    Native framework support

    In Practice

    What teams build with agent memory

    Research Agents

    Agents accumulate findings across multiple research sessions and surface relevant prior work when tackling new questions.

    Customer Support Bots

    Support agents remember past interactions with a customer and personalize responses based on conversation history.

    Coding Assistants

    Code agents store architectural decisions and past debugging outcomes so they apply lessons learned to new problems.

    Multi-agent Pipelines

    A crew of specialized agents shares memory, researcher, analyst, and writer agents collaborate through a common memory store.