ZeroDB
Memory, vectors, events, and files — one API for AI agents that remember. The Mem0 alternative built for production agents.
npx zerodb-cli initUsed by developers from leading tech companies and universities
One Platform. The Whole Agent Data Layer.
Memory, vectors, tables, a temporal knowledge graph, free embeddings, RLHF, agent observability, batch ops, and multimodal media — one API, cloud or local. Everything your agents need to remember, reason, and act.
ZeroMemory — Cognitive Agent Memory
Working memory, episodic recall, and semantic memory layers with automatic consolidation, decay scoring, and entity relationship graphs.
Vector Search with pgvector HNSW
Sub-millisecond semantic search backed by pgvector HNSW indexes. Embeddings included — no OpenAI key required.
NoSQL Tables + Event Streams
Schema-free document tables and real-time event streaming alongside vector storage — one API for your full agent data layer.
File Storage (S3-Compatible)
Agent-native object storage for documents, images, and artifacts. Multi-tenant scoping with fine-grained access controls.
MCP Server Integration
10-tool memory MCP server and 83-tool full database server. Works with Cursor, VS Code, and AINative Studio IDE. Hosted remote MCP available — no local install needed.
W3C Knowledge Graph + SPARQL
Your memory is a real RDF knowledge graph: query it with SPARQL 1.1, validate with SHACL, reason with OWL, and dereference every entity at a stable IRI. Bi-temporal edges, PROV-O provenance, GraphRAG hybrid search, and lossless Vault-LD export.
ZeroDB Functions
Event-triggered serverless functions that fire on data changes — automatically embed vectors the moment data is written. Sandboxed execution (Python / JS) benchmarked at 18ms P50, 112 RPS. Wire your own hooks to react to every write.
RLHF & Preference Data API
Collect, review, and export human-feedback and preference data for fine-tuning — built into your database. Log agent interactions, score them, compare pairs, run single-review, and export training-ready datasets (JSON, CSV, Parquet, OpenAI fine-tune format). The feedback loop that makes your agents smarter, with no separate labeling tool.
Bi-Temporal Graph — Time-Travel Your Knowledge
Every fact carries valid_from / valid_until. Query the graph as it was at any point in time with as-of queries, see how facts changed, and detect + resolve contradictions when new information conflicts with old. Temporal reasoning most vector databases simply don’t have.
Ontology Inference & Entity Resolution
ZeroDB learns the shape of your data — auto-infers entity types and relationship predicates, suggests constraints, resolves aliases to canonical entities, and builds entity timelines. A knowledge graph that improves itself as your agents write to it.
Agent Observability & Logs
Built-in monitoring for AI agents: active-agent metrics, task completion and error rates, average duration, recent-activity feeds, live execution traces, and per-session debugging. See what your agents actually did — and why they failed.
Batch Operations at Scale
High-throughput bulk APIs: 50 queries per batch (parallel, MongoDB-style filters), 500 memory records per atomic upsert with auto-embedding, 500-record bulk update/delete, and 1,000+ events/sec batch publishing. Built for agent workloads, not one-row-at-a-time.
Local ↔ Cloud Sync — No Lock-In
Develop offline with ZeroDB Local (`pip install zerodb-local`, `zerodb serve`), then bundle and sync your full project — vectors, tables, memory, events, files — to the cloud, or export it out cleanly. Same API everywhere. Your data is yours.
TurboQuant Vector Compression
Shrink your vector footprint with TurboQuant (PolarQuant + QJL, based on Google’s ICLR 2026 research) while preserving cosine similarity — sub-millisecond search at a fraction of the storage. Fast retrieval that scales with your agent’s memory.
Multimodal Memory — Video & Media
Your agents remember what they see. Automatic video frame extraction, visual embeddings, and scene detection, plus multi-format media transcoding — memory that spans text, images, and video, searchable by meaning.
Need a dedicated database? Add managed Postgres.
Beyond the built-in ZeroDB data layer, spin up your own fully-managed, dedicated PostgreSQL instance (v15–18) in seconds — scale up or down anytime. Priced per tier, billed monthly. A separate add-on, not part of the core plan.
micro-1
$15/mo
1 vCPU · 1 GB RAM · 10 GB
standard-2
$29/mo
25 GB storage
standard-4
$59/mo
50 GB storage
performance-8
$119/mo
100 GB storage
performance-16
$249/mo
high-performance
PostgreSQL 15, 16, 17 & 18 · scale between tiers anytime · billed monthly.
Get Started in 30 Seconds
Get a working database in one command. Start free with a 3-day trial, then $5/mo (Hobbyist). Python, Node.js, LangChain, LlamaIndex — pick your stack.
Instant Setup (3-day free trial)
npx zerodb-cli initPython SDK
pip install zerodb-mcpLangChain
pip install langchain-zerodbLlamaIndex
pip install llama-index-vector-stores-zerodbMCP Server (Agent Memory)
npm i ainative-zerodb-memory-mcp- Instant database — start free with a 3-day trial, then $5/mo (Hobbyist)
- Embeddings included — BAAI/bge models, no OpenAI costs
- Sub-millisecond search with HNSW indexes
- MCP servers for Claude Code, Cursor, VS Code, Windsurf
- LangChain + LlamaIndex integrations
# Start free with a 3-day trial, then $5/mo (Hobbyist)
curl -X POST https://api.ainative.studio/api/v1/public/instant-db
# Or use the CLI
npx zerodb-cli init
# Python — LangChain integration
from langchain_zerodb import ZeroDBVectorStore
store = ZeroDBVectorStore(
api_key="your-api-key",
project_id="your-project-id",
)
# Add documents (embeddings included — no OpenAI key)
store.add_texts(["ZeroDB is fast", "Semantic search"])
# Search by meaning
results = store.similarity_search("fast database", k=5)
# LlamaIndex integration
from llama_index_zerodb import ZeroDBVectorStore
from llama_index.core import VectorStoreIndex
index = VectorStoreIndex.from_vector_store(store)
response = index.as_query_engine().query("What is ZeroDB?")Deep Dive
Why ZeroDB is built for AI agents
See the six requirements every agent-ready vector database must meet, why generic vector stores break down for agent workloads, and how ZeroDB solves them — with code examples and a side-by-side comparison.
Read: Vector Database for AI AgentsZeroTime — Browser Agent
Give your agents eyes with Browser Agent MCP
Extract structured data from any website and store it directly in ZeroDB. Enrich CRM leads, monitor competitors, pull invoice data — all via npx @ainative/browser-mcp.
Give Your Agents Persistent Memory
One API for memory, vectors, events, and files. Start free with a 3-day trial, then $5/mo (Hobbyist).
Explore ZeroDB
One platform, one API key — every capability has a home. Dive into the piece you need.
Data layer
Vectors
pgvector HNSW semantic search for agents.
Embeddings API
Cheap, fast embeddings — embed-and-store + search.
Tables / NoSQL
JSON documents with MongoDB-style queries.
File Storage
S3-compatible object storage per project.
Events
Publish, subscribe, and trigger functions.
Dedicated Postgres
A managed Postgres per project with pgvector.
Memory & reasoning
Agent Memory
Remember, recall, forget — persistent cognitive memory.
Knowledge Graph
W3C RDF entities, relationships, and reasoning.
Knowledge Graph API
SPARQL 1.1, traversal, ontology, and GraphRAG.
RLHF / Preference Data
Collect feedback and export DPO/SFT datasets.
Lakehouse
DuckDB + Parquet analytics on your data.
