Nexus
AI-Native Filesystem for Building Intelligent Agents
Store files. Remember context. Control access. Scale instantly.
Give your AI agents a filesystem that thinksβwith memory, permissions, and learning built in.
β‘ What is Nexus? (10-second pitch)¶
Nexus is a programmable filesystem for AI agents that combines:
- π File Storage - Store and version files (local, S3, GCS)
- π§ Agent Memory - Agents remember and learn across sessions
- π Smart Permissions - Fine-grained access control (ReBAC/Zanzibar-style)
- π Semantic Search - Find anything with natural language
- π ZeroβProduction - Start local (
pip install), scale without code changes
One API. Multiple backends. Built for AI agents.
Quick Start in 30 Seconds¶
# Start server with authentication
nexus serve --host 0.0.0.0 --port 2026
# Connect from Python
import nexus
nx = nexus.connect(
remote_url="http://localhost:2026",
api_key="your-api-key"
)
nx.write("/workspace/hello.txt", b"Remote write!")
π Learn more: Administration & Operations (25 min)
π― Is Nexus For Me?¶
Answer 2 quick questions to find your path:
1. What are you building?¶
Perfect fit! Nexus is built specifically for AI agents.
β You need: Agent memory, persistent storage, multi-agent collaboration π Start here: AI Agent Memory (15 min)
Use cases: - Multi-agent teams (CrewAI, LangGraph workflows) - Chatbots with long-term memory across sessions - Code assistants that learn from your codebase - Research agents that build knowledge over time
Frameworks supported: CrewAI, LangGraph, Claude SDK, OpenAI Agents, Google ADK
Excellent choice! LLM-powered reading and semantic search built-in.
β You need: Document search, AI-powered Q&A, content processing π Start here: Document Q&A System (10 min)
Use cases: - Knowledge bases with natural language search - Document processing pipelines with batch uploads - Content analysis with LLM-powered insights - RAG systems with semantic search
Great choice! Built-in multi-tenancy and fine-grained permissions.
β You need: Workspace isolation, ReBAC permissions, audit trails π Start here: Multi-Tenant SaaS (30 min)
Use cases: - Collaboration platforms with team workspaces - Enterprise apps with tenant isolation - Data platforms with row-level access control - AI-powered SaaS with agent workspace sharing
Perfect! Event-driven workflows and plugin extensibility.
β You need: Automated pipelines, custom integrations, event triggers π Start here: Workflow Automation (15 min)
Use cases: - Data pipelines with automatic processing - File-based triggers (e.g., invoice β extract β store) - Custom integrations via plugin system - ML experiment tracking with versioned checkpoints
Excellent! Package and share AI capabilities as reusable skills.
β You need: Skill discovery, documentation-to-skill conversion, team sharing π Start here: Skills Management (20 min)
Use cases: - Team skill libraries from internal/external documentation - Auto-generate skills from any docs URL - Skill discovery across agent/tenant/system tiers - Knowledge sharing with ReBAC permissions - AI agent enhancement with contextual skills
2. What's your experience level?¶
Recommended Path:
- Quick Start (30 sec) - Copy-paste working code
- Core Concepts (5 min) - Understand the mental model
- Choose Your Learning Path - Pick based on use case
π‘ Tip: Start with embedded mode, switch to server later (same code!)
Jump straight to examples:
π Reference: Complete API Docs
Advanced guides:
- Administration & Operations (25 min) - User management, API keys, monitoring
- Multi-Backend Setup (20 min) - Configure S3, GCS, databases
- Building Plugins (30 min) - Extend Nexus with custom functionality
- Security Hardening - Production security best practices
- Performance Tuning - Optimize for scale
Still not sure? Check out Why Nexus? for detailed comparisons with S3, Vector DBs, and traditional filesystems.
How It Works¶
%%{init: {'theme':'base', 'themeVariables': { 'primaryColor':'#e3f2fd','primaryTextColor':'#1a237e','primaryBorderColor':'#5C6BC0','lineColor':'#AB47BC','secondaryColor':'#fce4ec','tertiaryColor':'#fff3e0','fontSize':'14px'}}}%%
graph TB
subgraph agents[" π€ AI Agents "]
agent1["Agent A<br/>(GPT-4)"]
agent2["Agent B<br/>(Claude)"]
agent3["Agent C<br/>(Custom)"]
end
subgraph vfs[" π Nexus Virtual File System "]
api["Unified VFS API<br/>read() write() list() search()"]
memory["πΎ Memory API<br/>Persistent learning & context"]
rebac["π ReBAC Permissions<br/>Backend-aware object types"]
version["π¦ Versioning<br/>Snapshots & time-travel"]
router["Smart Router<br/>Path β Backend + Object Type"]
end
subgraph backends[" πΎ Storage & Data Backends "]
subgraph storage[" File Storage "]
local["Local Filesystem<br/>object: file"]
gcs["Cloud Storage<br/>object: file"]
end
subgraph data[" Data Sources "]
postgres["PostgreSQL<br/>object: postgres:table/row"]
redis["Redis<br/>object: redis:instance/key"]
mongo["MongoDB<br/>object: mongo:collection/doc"]
end
end
agent1 -.->|"write('/workspace/data.json')"| api
agent2 -.->|"read('/db/public/users')"| api
agent3 -.->|"memory.store('learned_fact')"| memory
api --> rebac
memory --> rebac
rebac <-->|"Check with object type"| router
rebac -->|"β Allowed"| version
version --> router
router -->|"File operations"| local
router -->|"File operations"| gcs
router -->|"Queries as files"| postgres
router -->|"KV as files"| redis
router -->|"Documents as files"| mongo
style agents fill:#e3f2fd,stroke:#5C6BC0,stroke-width:2px,color:#1a237e
style vfs fill:#f3e5f5,stroke:#AB47BC,stroke-width:2px,color:#4a148c
style backends fill:#fff3e0,stroke:#FF7043,stroke-width:2px,color:#e65100
style storage fill:#e8f5e9,stroke:#4CAF50,stroke-width:1px
style data fill:#e1f5fe,stroke:#0288D1,stroke-width:1px
style api fill:#5C6BC0,stroke:#3949AB,stroke-width:2px,color:#fff
style memory fill:#AB47BC,stroke:#7B1FA2,stroke-width:2px,color:#fff
style rebac fill:#EC407A,stroke:#C2185B,stroke-width:2px,color:#fff
style version fill:#66BB6A,stroke:#388E3C,stroke-width:2px,color:#fff
style router fill:#42A5F5,stroke:#1976D2,stroke-width:2px,color:#fff Backend Abstraction:
Nexus presents everything as files to users, while backends provide appropriate object types for permission control:
- File Storage (Local, GCS, S3): Standard file objects
- Databases (PostgreSQL, Redis, MongoDB): Backend-specific objects (tables, keys, documents)
- Unified Interface: All accessed through the same VFS API (read/write/list)
- Fine-Grained Permissions: ReBAC uses backend-appropriate object types (e.g., grant access to a PostgreSQL schema vs. individual rows)
π― One API¶
Agents use simple file operations, regardless of where data lives. No cloud SDKs to learn.
π Built-in Security¶
Every operation checks permissions automatically. Google Zanzibar-style ReBAC included.
πΎ Agent Memory¶
Persistent learning across sessions. Agents remember context and improve automatically.
π¦ Versioning¶
Time-travel debugging with snapshots. Roll back to any point in history instantly.
π Backend Flexibility¶
Switch from local to cloud without changing agent code. Zero vendor lock-in.
π Production Ready¶
Multi-tenancy, workspace isolation, and complete audit trails out of the box.
β‘ Workflow Automation¶
Event-driven workflows fire automatically on file operations. No manual triggering needed.
π§ Core Concepts¶
New to Nexus? Start here to understand how it works:
-
What is Nexus?
Mental model and architecture overview. Understand the three-layer system and key design principles.
-
Memory System
How agents remember context across sessions. Learn about namespaces, identity-based memory, and persistence.
-
Agent Permissions
How agents authenticate and inherit permissions from users. ReBAC for multi-agent collaboration.
-
ReBAC Explained
Deep dive into Relationship-Based Access Control. Zanzibar-style permissions with tuples and expansion.
-
Content-Addressable Storage
How CAS works with SHA-256 hashing. Automatic deduplication and immutable version history.
-
Learning Loops
Algorithm-agnostic learning primitives. Trajectories, reflection, playbooks, and feedback collection.
-
Workflows & Triggers
Event-driven automation with file patterns. Auto-process documents, validate data, and send webhooks.
π Choose Your Learning Path¶
Progressive learning from beginner to advanced. Pick based on your use case and experience level.
π± Beginner Paths (5-15 minutes)¶
-
π Simple File Storage
Master basic read/write operations with local and cloud backends.
β±οΈ 5 min | π‘ Difficulty: Easy
-
π Document Q&A System
Build semantic search and LLM-powered document reading.
β±οΈ 10 min | π‘ Difficulty: Easy
-
π§ AI Agent Memory
Add persistent memory and learning to your agents.
β±οΈ 15 min | π‘ Difficulty: Medium
-
π Workflow Automation
Build event-driven workflows that trigger automatically.
β±οΈ 15 min | π‘ Difficulty: Medium
π Intermediate Paths (15-25 minutes)¶
-
π₯ Team Collaboration
Multi-user access with permissions and workspace isolation.
β±οΈ 20 min | π‘ Difficulty: Medium
-
π€ Agent Framework Integration
Connect LangGraph, CrewAI, Claude SDK, and OpenAI Agents.
β±οΈ 20 min | π‘ Difficulty: Medium
-
πΎ Multi-Backend Storage
Mount and federate S3, GCS, and database backends.
β±οΈ 20 min | π‘ Difficulty: Medium
-
π§ Agent Memory & Learning
Build agents that learn and improve over time.
β±οΈ 20 min | π‘ Difficulty: Medium
βοΈ Advanced Paths (25-30+ minutes)¶
-
π’ Multi-Tenant SaaS
Build production SaaS with complete tenant isolation.
β±οΈ 30 min | π‘ Difficulty: Hard
-
β‘ Skills Management
Create, version, and publish reusable agent skills.
β±οΈ 15 min | π‘ Difficulty: Medium
-
π§ Administration & Operations π
User management, API keys, monitoring, and backup.
β±οΈ 25 min | π‘ Difficulty: Hard | π‘ Server Mode
-
π Building Plugins π
Extend Nexus with custom functionality and lifecycle hooks.
β±οΈ 30 min | π‘ Difficulty: Hard
Not sure where to start? Use the Is Nexus For Me? quiz above to find your recommended path.
What Makes Nexus Different?¶
For AI Agent Developers¶
| Feature | Vector DB (Pinecone) | Object Storage (S3) | Traditional FS | Nexus |
|---|---|---|---|---|
| Agent Memory | π‘ Embeddings only | β | β | β Memory + Learning |
| File Operations | β | π‘ Object-only | β | β Full POSIX-like API |
| Semantic Search | β | β | β | β Vector + metadata |
| Multi-Agent Collaboration | β | β | β | β Shared workspaces |
| Permissions & ReBAC | π‘ Basic | π‘ IAM policies | π‘ UNIX perms | β Zanzibar-style ReBAC |
| Multi-Tenancy | π‘ Collections | π‘ Manual buckets | β | β Native isolation |
| Event-Driven Workflows | β | π‘ S3 notifications | β | β Auto-trigger on writes |
| Time Travel & Versioning | β | π‘ Versioning only | β | β Full history + diffs |
| Database Access as Files | β | β | β | β PostgreSQL, Redis, MongoDB |
| LLM Document Reading | β | β | β | β Query docs with citations |
Why AI Developers Choose Nexus:
- π€ Multi-Agent Native: Built-in workspace sharing and collaboration for CrewAI, LangGraph teams
- π§ Memory + Storage Combined: Not just embeddingsβfull agent memory with automatic consolidation
- π Event-Driven Automation: Workflows trigger automatically when agents write files (no manual event firing)
- ποΈ Unified Data Access: Access files, databases (PostgreSQL/Redis/MongoDB), and vectors through one API
- π Production-Ready Security: Fine-grained ReBAC permissions with multi-tenancy out of the box
- π LLM-Powered Intelligence: Ask questions about stored documents, get AI answers with citations
- β‘ Framework Integrations: First-class support for CrewAI, LangGraph, Claude SDK, OpenAI Agents, Google ADK
π Need Help?¶
-
π Quickstart Guide
Step-by-step setup in under 10 minutes
-
π§ Troubleshooting
Common issues and solutions
-
π‘ Core Concepts
Understand how Nexus works
-
π Glossary
Complete A-Z reference: ReBAC, CAS, ACE, and all Nexus terminology