The filesystem & context plane for AI agents — unified virtual filesystem with versioning, semantic search, persistent memory, ReBAC authorization, MCP integration, and federated deployment.
Nexus is a filesystem and context infrastructure designed for AI agents, built on a three-layer architecture: Bricks → Kernel → Drivers. The immutable Kernel provides a POSIX-style virtual filesystem, CAS deduplication, and three-phase writes; pluggable Bricks modules cover versioning, snapshots, hybrid search (Zoekt + pgvector), persistent memory, Zanzibar-based ReBAC delegation, bidirectional MCP integration, workflow automation, governance guardrails, and metered billing across 25+ modules; swappable Drivers support redb (embedded), PostgreSQL, S3/GCS, Dragonfly, and other backends.
Three deployment modes: embedded (zero infra, nexus.connect()), shared daemon (team multi-agent collaboration), and federated (cross-datacenter Raft consensus + mTLS TOFU). Exposes 30+ MCP tools and integrates out-of-the-box with Claude Agent SDK, OpenAI Agents, LangGraph, CrewAI, Google ADK, and E2B. Built-in cold storage tiering reduces cold data costs by ~80%. Kernel is Python + Rust (compiled via maturin), cross-platform (Linux/macOS/Windows), Docker multi-arch (amd64 + arm64).
Four Storage Pillars
| Pillar | Interface | Capabilities | Required |
|---|---|---|---|
| Metastore | MetastoreABC | Ordered KV, CAS, prefix scan, optional Raft | ✅ Only kernel init param |
| ObjectStore | ObjectStoreABC | Streaming Blob I/O, PB-scale | Dynamic mount |
| RecordStore | RecordStoreABC | Relational ACID, JOIN, vector search | Optional |
| CacheStore | CacheStoreABC | Ephemeral KV, Pub/Sub, TTL | Optional |
Quick Start
Docker mode (recommended):
pip install nexus-ai-fs
nexus init --preset demo
nexus up
eval $(nexus env)
Embedded mode:
import asyncio, nexus
async def main():
nx = await nexus.connect(config={"data_dir": "./my-data"})
await nx.write("/notes/meeting.md", b"# Q3 Planning\n- Ship Nexus 1.0")
print((await nx.read("/notes/meeting.md")).decode())
nx.close()
asyncio.run(main())
Package Distribution
nexus-ai-fs(PyPI): Full package with CLI + SDKnexus-fs(PyPI): Slim package@nexus-ai-fs/tui(npm): Terminal UI
Unconfirmed: Nexi Labs team background not publicly disclosed; v1.0 timeline unclear; Zanzibar ReBAC implementation completeness and Governance/Pay module maturity (GA vs preview) unverified; nexus-fuse subproject functionality undocumented; Python 3.14+ hard requirement may be a barrier for early adopters.