> For the complete documentation index, see [llms.txt](https://docs.nerve-protocol.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.nerve-protocol.com/overview/nerve-protocol-overview.md).

# Nerve Protocol Overview

**Nerve Protocol** is a privacy-native AI operating layer built for operators who require cryptographic guarantees over their data, models, and execution environment. Every workflow runs inside attested Trusted Execution Environments. Intent compiles into signed mission graphs. Outputs stream back encrypted. The host at no point has access to plaintext.

Nerve is not a wrapper around existing cloud AI. It is a sovereign infrastructure layer: you hold the keys, the weights are open, and every action produces verifiable proof.

***

## Design Principles

**Hardware-enforced isolation, not policy.** Data and inference are protected at the silicon level via Trusted Execution Environments. Compliance is structural — it cannot be misconfigured away or overridden by a compromised application layer.

**Operator-held key custody.** Signing and decryption keys are generated inside the enclave and never leave it. Even Nerve's own orchestration infrastructure cannot extract raw key material.

**Verifiable execution over trusted execution.** Every enclave used in a mission publishes a remote attestation quote before it receives work. Completion produces zero-knowledge receipts anchored to an on-chain Coordination Ledger — verifiable by any party without access to mission content.

**Composable without compromise.** Operators can connect external systems, deploy custom agents, and extend the protocol without weakening the cryptographic perimeter. Each new integration inherits purpose-bound policy capsules.

**Ephemeral by default.** Sessions dissolve on completion. Enclave memory is wiped, capsules dissolve, and only cryptographic attestation hashes persist to the ledger.

***

## Core Components

**Secure Data Connectors** — Circuit-level TEE-resident modules that retrieve, normalize, and encrypt data from external sources before it contacts the host OS. They interface with SaaS APIs (Uber, Amazon, Slack, Notion, Google Workspace), on-chain protocols, and air-gapped hardware. OAuth tokens and API credentials are sealed inside the enclave — the host never holds them in plaintext. See [Secure Data Connectors](/overview/data-integrators.md).

**On-Device LLMs — Myelin** — A 4B-parameter edge model with native audio and video processing, a 128K context window, and offline-capable execution on consumer hardware. Myelin is the inference layer for on-device Personal AI. Queries never leave the device or enclave. See [On-Device LLMs and Myelin](/overview/on-device-llms.md).

**Soma** — The 32B uncensored flagship model. Dense architecture, 256K context, fully open weights under Apache 2.0. Soma runs inside a sealed TEE on the Nerve grid or on operator-controlled hardware. Zero refusals. The host never sees prompts or outputs. See [Soma](/overview/soma.md).

**Enclave Integrity Monitors** — Specialized validation models co-located inside TEEs that continuously audit agent behavior, verify policy capsule adherence, detect anomalous execution patterns, and emit cryptographic compliance receipts. They cannot read mission content — they evaluate behavioral signals. See [Enclave Integrity Monitors](/overview/guardian-models.md).

**Signal Lattice** — An encrypted analytics fabric that converts raw data events into actionable intelligence streams. Signals are rendered exclusively within authenticated operator sessions and are never stored in plaintext.

**Executor Swarm** — Autonomous agents that run end-to-end inside TEEs under strict policy guardrails. Executors can schedule calls, prepare documents, reconcile transactions, and orchestrate cross-system workflows. Each step requires a fresh attestation check before proceeding.

***

## Operator Journey

A typical Nerve Protocol session follows a deterministic sequence:

1. **Authenticate.** The operator presents hardware-backed keys. The console boots only after the enclave passes remote attestation.
2. **Compile intent.** Natural language input is converted into a deterministic mission graph. The compiled graph is hashed and committed to the audit trail before any agent receives it.
3. **Wire the grid.** External systems are connected via [Secure Data Connectors](/overview/data-integrators.md) under purpose-bound policy capsules. Each integration declares scope, retention window, and expiration before data flows.
4. **Deploy executors.** Autonomous agents are assigned to mission steps. Enclave Integrity Monitors validate each stage boundary and emit compliance receipts.
5. **Extract signal.** Encrypted insights are rendered inside the operator's active session. Zero-knowledge proofs of mission completion are generated and available for selective disclosure.
6. **Close session.** On mission completion or explicit close, enclave memory is wiped, session capsules dissolve, and only attestation hashes persist.

***

## Protocol Layers

Nerve Protocol is organized into five layers, each handling a distinct aspect of the execution environment:

| Layer                   | Function                                                                                         |
| ----------------------- | ------------------------------------------------------------------------------------------------ |
| **Edge Capsules**       | Client-side TEEs that host the Nerve Console, hold decryption keys, and render output locally    |
| **Enclave Mesh**        | Elastic pools of hardware enclaves running executors, connectors, and Enclave Integrity Monitors |
| **Signal Lattice**      | Privacy-preserving analytics fabric converting raw events to encrypted intelligence streams      |
| **Coordination Ledger** | On-chain log of mission hashes, policy states, and zero-knowledge compliance proofs              |
| **Agent Exchange**      | Marketplace where vetted agents and connectors publish capabilities and reputation scores        |

For a deep dive into each layer, see the [Infrastructure Overview](/architecture/nerve-protocol-infrastructure-overview.md).


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.nerve-protocol.com/overview/nerve-protocol-overview.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.