Skip to main content
Agent Onboarding

Give Your Agent
a Persistent Brain

ZeroMemory is the persistent knowledge layer for AI agents. Three API calls — remember, recall, forget — give your agent a brain that persists across sessions.

100% Recall@1·LongMemEval (ICLR 2025)
60-second quickstartAPI Reference

What is ZeroMemory?

ZeroMemory is the persistent knowledge layer for AI agents. Three API calls — remember, recall, forget — give your agent a brain that persists across sessions. No infrastructure to manage, no vector databases to tune. You write facts in, you query them with natural language, and you delete them when they expire. ZeroMemory handles embedding, indexing, consolidation, and retrieval ranking so your agent can focus on reasoning.

remember

Persist a fact

recall

Query by meaning

forget

Remove a memory

When to Use It

Your situationUse this

Building a chatbot that remembers users across sessions

ZeroMemory

Need RAG with full vector search and metadata filtering

ZeroDB vectors

Orchestrating multi-agent workflows with shared state

Agent Swarm + shared memory

Connecting memory tools to Claude / Cursor / Windsurf

MCP hosting

Quickstart (60 seconds)

Step 1 — Install the MCP package

Terminalbash
npm i ainative-zerodb-memory-mcp

Step 2 — Add to your MCP config (claude_desktop_config.json or .cursor/mcp.json)

MCP configjson
{
  "mcpServers": {
    "zerodb-memory": {
      "command": "npx",
      "args": ["-y", "ainative-zerodb-memory-mcp"],
      "env": {
        "AINATIVE_API_KEY": "your-api-key",
        "AINATIVE_PROJECT_ID": "your-project-id"
      }
    }
  }
}

Get your API key from the dashboard. The MCP server exposes six tools: zerodb_store_memory, zerodb_search_memory, zerodb_semantic_search, zerodb_get_context, zerodb_embed_text, zerodb_clear_session.

Core Operations

Every example below uses the REST API directly. The Python SDK mirrors the same method names.

remember— persist a fact

Store any text as a memory. AINative embeds it, indexes it, and makes it retrievable by meaning.

curlbash
curl -X POST https://api.ainative.studio/api/v1/public/memory/v2/remember \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "User prefers TypeScript over Python for backend work",
    "session_id": "user-123",
    "tags": ["preference", "stack"]
  }'
Python SDKpython
await memory.remember(
    content="User prefers TypeScript over Python for backend work",
    session_id="user-123",
    tags=["preference", "stack"]
)

recall— query by meaning

Ask a natural-language question. ZeroMemory returns ranked memories by semantic similarity.

curlbash
curl -X POST https://api.ainative.studio/api/v1/public/memory/v2/recall \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "What language does this user prefer?",
    "session_id": "user-123",
    "limit": 5
  }'
Python SDKpython
results = await memory.recall(
    query="What language does this user prefer?",
    session_id="user-123",
    limit=5
)
for item in results:
    print(item.content, item.score)

forget— remove a memory

Explicitly delete a memory by ID. Use this for GDPR right-to-erasure flows or stale fact removal.

curlbash
curl -X POST https://api.ainative.studio/api/v1/public/memory/v2/forget \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "memory_id": "mem_abc123"
  }'
Python SDKpython
await memory.forget(memory_id="mem_abc123")

Full Python example

Python — all three operationspython
from ainative import ZeroMemory

memory = ZeroMemory(api_key="YOUR_API_KEY")

# Remember something
await memory.remember(
    content="User prefers TypeScript over Python for backend work",
    session_id="user-123",
    tags=["preference", "stack"]
)

# Recall relevant memories
results = await memory.recall(
    query="What language does this user prefer?",
    session_id="user-123",
    limit=5
)
for item in results:
    print(item.content, item.score)

# Forget a specific memory
await memory.forget(memory_id="mem_abc123")

Memory Patterns

ZeroMemory supports three complementary memory modes. Use them together for production agents.

Episodic

Conversation turns and chat history

"User asked about pricing on 2026-03-20"

Semantic

Consolidated insights from reflection

"User is an engineer at a Series B startup"

Working

Short-term scratchpad for the current task

"Current draft: feature spec for auth module"

Pro tip: Call /reflect after a session to automatically consolidate episodic memories into semantic ones. ZeroMemory handles the summarisation and deduplication.

Benchmark

100%

Recall@1

LongMemEval benchmark (ICLR 2025)

ZeroMemory achieves 100% Recall@1 and 94% QA accuracy on LongMemEval — outperforming the GPT-4o oracle baseline. LongMemEval is the most rigorous publicly available benchmark for long-term conversational memory retrieval.

Read the benchmark post

Next Steps

API Reference

Complete endpoint docs for all ZeroMemory operations, parameters, and response shapes.

ZeroDB Product Page

Explore vector storage, semantic search, and file storage — the full ZeroDB platform.

Blog

Deep dives on memory architectures, benchmark methodology, and agent patterns.

Download IDE

AINative Studio IDE — multi-agent development with ZeroMemory built in.

Ready to build?

Free plan includes 10,000 memory operations per month. No credit card required.

Start for freeBack to Getting Started