AGENT SUBSTRATE  ·  OPEN SOURCE

FUSE is all agents
need

NervaFS is the substrate layer for agent systems. Keep storage, identity, policy, tools, and mounted capabilities behind one filesystem contract, then let runtimes and orchestration bind above it.

nervafs — bash
# Bind a runtime to the substrate
$ nervafs mount agent-001
Mounted at /agents/agent-001/
Substrate ready · runtime-agnostic contract available
 
$ ls /agents/agent-001/
identity/   memory/    skills/
policy/     logs/     snapshots/
data/       channels/   rag/   tools/
 
$
scroll to prove it
// WHY FUSE

Agent systems need separation,
not more bundling.

Most agent stacks collapse storage, execution, and coordination into one moving target. NervaFS argues for a different shape: keep Substrate, Runtime, and Orchestration separate, then connect them through one filesystem-native contract.

What NervaFS is

A substrate layer that makes identity, memory, policy, logs, tools, channels, and mounted systems inspectable and governable without forcing a runtime or orchestration opinion.

Substrate holds persistent agent state and mounted capabilities.
Runtime binds models and tool execution onto that substrate.
Orchestration coordinates flows above both, instead of swallowing them.
AGENT
PATH
CORE SUBSTRATE
/identity
/persona
CONTENTS
ARCHITECTURE
Agent Orchestration
Planning · routing · multi-agent workflows
Agent Runtime
Execution · LLM calls · tool use
Agent Substrate (Storage)
Identity · memory · skills · policy · logs
OURS
Coresubstrate
/identity
/memory
/skills
/policy
/logs
LAYER DETAIL
Agent Orchestration
Coordinates multiple agents to accomplish complex tasks. Handles intent parsing, task delegation, and result aggregation across the agent graph.
COMMERCIAL
OPEN SOURCE
NervaFS is our open-source answer to the missing infrastructure layer — the storage substrate every agent runtime depends on, but nobody has solved.
// WHAT SEPARATION BUYS YOU

What the substrate layer gives you

This is not a feature buffet. It is the operational upside of keeping the agent substrate stable while runtimes and orchestration evolve above it.

🗂
Unified Filesystem Abstraction
Every resource — memory, tools, data, channels — becomes a file path. Agents use familiar read, write, ls semantics.
🔐
Fine-grained RBAC
Visibility-level permission control per agent, per path, per role. Sensitive fields filtered at the FUSE layer before agents see them.
🔌
Plugin Architecture
Mount any backend — databases, APIs, vector stores, Vault secrets — as a virtual path via hot-pluggable modular plugins.
📦
Agent Bundle Packaging
Pack identity, memory, skills, and policy into a portable .agent_bundle. Deploy on any runtime in seconds, any environment.
🔍
Full Audit Trail
Every file operation logged. Dangerous actions intercepted. Real-time admin alerts. 100% audit coverage guaranteed.
Runtime-Agnostic
Works with Claude Code, OpenClaw, Hermes, Codex, and any future runtime. NervaFS is the infrastructure layer beneath them all.
// EXTENSIBILITY

Extend the substrate
without collapsing the model

Plugins are how NervaFS grows without turning back into a bundled stack. Mount systems, replace stores, and package capabilities while keeping the same substrate contract.

Tool / CLI Plugin
Wrap any external API or executable as a standard CLI tool. NervaFS registers it in /tools/ and exposes it to agents with a man page for self-discovery.
Slack notifierSQL executorImage processorCustom LLM
plugin.yaml
type: tool
name: my-tool
mount: /tools/my-tool
entrypoint: ./bin/my-tool
interface:
  stdin: json
  stdout: json
  exitcode: standard
man: ./docs/man.md
01
Define your plugin
Create a plugin.yaml declaring type, mount path, entrypoint, and interface contract.
02
Implement the interface
Write your handler in any language. NervaFS communicates via stdin/stdout JSON or FUSE hook scripts.
03
Install & mount
Run nervafs plugin install ./my-plugin. The path appears instantly in every agent's filesystem.
04
Agents discover it
Agents read your man doc or .readme.skill and start using the new capability immediately.
// JOIN THE THESIS

Choose your path in

The page should teach the idea first, then let visitors self-select. Use docs if you want to inspect, use the waitlist if you want updates, and use Discord if you want to discuss the architecture now.

INFRA TEAM
Explore the substrate as platform infrastructure
Use this if you are deciding whether storage / compute separation belongs in your internal agent stack.
BUILDER
Read the docs, then join the waitlist
Start with the architecture and capability model, then signal that you want access when the implementation matures.
DESIGN PARTNER
Pressure-test the architecture with us
If you are shaping a real internal agent platform, use this path to discuss design tradeoffs instead of just collecting updates.
HIGH-INTENT
Join the Discord and debate the model
This is the direct path for people who want to discuss the substrate/runtime/orchestration split in public as the project evolves.