AGiD — Auditable Agent Identity

Compliance infrastructure for AI agents. Cryptographic identity, tamper-proof audit trails, end-to-end encrypted communication, and verifiable on-chain memory.

Install as an OpenClaw plugin or run the full AGiD runtime with self-hosted messaging.

Architecture

CLI / Client
    ↓  Encrypted messages (ECDH)
MessageBox (self-hosted relay)
    ↓  Authenticated WebSocket
AGiD Gateway
    ↓  Plugin system (57 tools)
LLM Provider (Anthropic / Ollama / OpenAI-compatible)
    ↓  Tool execution
Identity Client (sign, pay, certify, prove)
    ↓  Encrypted response
MessageBox → Client

The gateway runs a native agent loop iterating between LLM calls and tool execution. Read-only tools execute in parallel. Wallet-mutating tools execute sequentially for signing safety. All 57 tools are registered through the modular plugin system.

Capabilities

• Cryptographic Identity — Every agent gets a unique key pair via the Identity Client. Hierarchical key derivation, identity certificates, and identity overlay resolution.
• 57 Tools across 11 plugins — identity, crypto, wallet, messaging, memory, audit, certificates, zero-knowledge proofs, deployment, runtime, file system, browser.
• Encrypted Memory — On-chain tokens with derived encryption keys. RLM-powered semantic search and deep reasoning.
• E2E Encrypted Messaging — Self-hosted MessageBox with ECDH encryption. Authenticated WebSocket for live communication. The relay never sees plaintext.
• Audit Trail — Per-session hash chains with Merkle root commitment on-chain. Workspace integrity verification against immutable anchors.
• Zero-Knowledge Proofs — Schnorr-based proofs for privileged communication verification, selective session revelation, and content commitments.
• Certificate System — Issue, verify, revoke, reveal, and exchange identity certificates.
• Multi-LLM Support — Anthropic Claude, Ollama (local), or any OpenAI-compatible endpoint.
• Infrastructure Deployment — Deploy and manage projects on Mandala Network nodes directly from agent tools.
• General-Purpose Execution — Shell commands, file operations, and headless browser automation.

OpenClaw Plugin

One install. 57 tools. Zero configuration.

openclaw plugin install @agid/openclaw-plugin

The Identity Client initializes on first use. Keys are generated and persisted locally. Every tool is immediately available.

openclaw config set agid.network mainnet
openclaw config set agid.storagePath /secure/path/identity.sqlite

AGiD Runtime

Full control. Self-hosted. Air-gappable.

Prerequisites

• Node.js >= 22.0.0
• Private key (generate with openssl rand -hex 32)
• Anthropic API key (or Ollama / OpenAI-compatible endpoint)

Install & Run

npm install
npm run build
npm run gateway

On first run, an interactive setup wizard walks through:

1. Private key generation or import
2. Network selection (mainnet / testnet)
3. LLM provider configuration
4. Certificate issuance
5. Workspace file creation

Configuration is persisted to ~/.agidentity/.

Manual Configuration

Create ~/.agidentity/.env:

AGENT_PRIVATE_KEY=<64-char-hex>
AGID_NETWORK=mainnet
ANTHROPIC_API_KEY=sk-your-key-here
TRUSTED_CERTIFIERS=<comma-separated-public-keys>

Environment Variables

Variable	Default	Description
AGENT_PRIVATE_KEY	required	64-char hex private key
ANTHROPIC_API_KEY	—	Anthropic API key
AGID_LLM_PROVIDER	anthropic	anthropic, ollama, or openai-compatible
AGID_MODEL	claude-sonnet-4-5-20250929	Model identifier
AGID_LLM_BASE_URL	—	Base URL for Ollama / OpenAI-compatible
AGID_LLM_API_KEY	—	API key for OpenAI-compatible
AGID_NETWORK	mainnet	mainnet or testnet
AGID_STORAGE_MODE	local	local or remote
AGID_WORKSPACE_PATH	~/.agidentity/workspace/	Workspace files directory
AGID_SESSIONS_PATH	~/.agidentity/sessions/	Session transcript directory
AGID_MAX_ITERATIONS	25	Max tool-use iterations per request
AGID_MAX_TOKENS	8192	Max tokens per LLM response
AGID_REQUIRE_CERTS	false	Reject messages from uncertified senders
AGID_ALLOW_UNAUTHENTICATED	false	Allow unauthenticated HTTP endpoints
MESSAGEBOX_HOST	https://messagebox.babbage.systems	MessageBox server URL
UHRP_STORAGE_URL	https://nanostore.babbage.systems	UHRP storage endpoint

Agent Tools

Identity (5 tools)

Tool	Description
agid_identity	Get agent's public key, network, and message box status
agid_balance	Check wallet balance in satoshis
agid_get_public_key	Derive protocol-specific keys via hierarchical key derivation
agid_get_height	Get current blockchain block height
agid_lookup_identity	Look up people on the identity overlay network

Crypto (5 tools)

Tool	Description
agid_sign	Sign messages to prove authorship
agid_encrypt	Encrypt data for secure storage or communication
agid_decrypt	Decrypt previously encrypted data
agid_wallet_client_request	Request user's Identity Client for cryptographic operations
agid_request_user_signature	Request user signature to prove user authorship

Wallet (7 tools)

Tool	Description
agid_create_action	Create transactions with outputs, baskets, and tags
agid_internalize_action	Accept incoming transactions
agid_list_outputs	List wallet outputs filtered by basket and tags
agid_send_payment	Send payments to another identity
agid_token_create	Create PushDrop tokens with arbitrary data fields
agid_token_list	List tokens from wallet baskets
agid_token_redeem	Redeem tokens to reclaim satoshis

Certificates (8 tools)

Tool	Description
agid_cert_issue	Issue identity certificates to other public keys
agid_cert_receive	Receive and store incoming certificates
agid_cert_list	List certificates in wallet
agid_cert_verify	Verify serialized certificates cryptographically
agid_cert_revoke	Revoke previously issued certificates
agid_cert_reveal	Publicly reveal selected certificate fields
agid_cert_check_revocation	Check on-chain revocation status
agid_cert_send	Send certificates via MessageBox

Zero-Knowledge Proofs (5 tools)

Tool	Description
agid_zkproof_privilege	Prove privileged communication without revealing content
agid_zkproof_verify	Verify a zero-knowledge proof
agid_zkproof_selective_reveal	Reveal one session's key without exposing others
agid_zkproof_commitment	Create tamper-evident content commitment anchored on-chain
agid_zkproof_verify_commitment	Verify content matches a previous commitment

Messaging (5 tools)

Tool	Description
agid_message_send	Send encrypted messages via MessageBox
agid_message_list	List messages in a box (auto-decrypted)
agid_message_ack	Acknowledge processed messages
agid_list_payments	List pending incoming payments
agid_accept_payment	Accept incoming payments

Memory (4 tools)

Tool	Description
agid_store_memory	Encrypt and store memories on-chain
agid_recall_memories	Recall with optional semantic search
shad_deep_recall	RLM deep reasoning over stored memories
shad_search_memories	Fast hybrid keyword + semantic search

Audit (2 tools)

Tool	Description
agid_verify_workspace	Verify workspace file integrity against on-chain anchor
agid_verify_session	Verify session anchor chain integrity

Deploy (8 tools)

Tool	Description
agid_mandala_create_project	Create a new project on a Mandala Node
agid_mandala_list_projects	List projects the agent has access to
agid_mandala_project_info	Get detailed project info
agid_mandala_deploy	Create deployment and get upload URL
agid_mandala_update_settings	Update project settings and environment
agid_mandala_project_logs	View project and resource logs
agid_mandala_manage_admins	Add, remove, or list project admins
agid_mandala_node_info	Get public node info

Runtime (2 tools)

Tool	Description
exec	Run shell commands with background process support
process	Manage background processes (poll, send-keys, submit)

File System (4 tools)

Tool	Description
read	Read a file with optional line range
write	Write or create a file
edit	Replace text in a file
apply_patch	Apply unified diff patches

Browser (1 tool)

Tool	Description
browser	Control headless Chromium (navigate, click, type, screenshot, evaluate)

Optimize (1 tool)

Tool	Description
agid_optimize_prompt	Evolutionary prompt optimization

Workspace Files

The agent's persona and context are defined by workspace files in ~/.agidentity/workspace/:

File	Purpose
SOUL.md	Core persona, behavioral rules, and values
IDENTITY.md	Agent's self-description and capabilities
TOOLS.md	Tool usage guidelines and preferences
MEMORY.md	Persistent memory context across sessions

These files are loaded into the system prompt on every request. Edit them to shape the agent's behavior.

API Endpoints

Endpoint	Auth	Description
GET /	No	Health check (status, public key, model, uptime)
GET /health	No	Health check
/identity/*	Mutual auth	Identity operations
/vault/*	Mutual auth	Vault storage operations
/team/*	Mutual auth	Team vault operations

All wallet operations are executed in-process via the plugin registry — no external API surface for wallet actions.

Docker

docker build -t agid .

docker run -d \
  -p 3000:3000 \
  -v agid-data:/data \
  -e AGENT_PRIVATE_KEY=<key> \
  -e ANTHROPIC_API_KEY=<key> \
  -e TRUSTED_CERTIFIERS=<keys> \
  agid

Project Structure

src/
├── plugins/         # Plugin system (types, registry, API, loader)
│   └── builtin/     # 11 builtin plugins (57 tools)
├── agent/           # Agent loop, prompt builder
│   └── tools/       # Legacy tool index (migrated to plugins)
├── audit/           # Signed audit trail, anchor chains
├── config/          # Configuration loading
├── encryption/      # Encryption helpers
├── gateway/         # MessageBox gateway, agent orchestration
├── identity/        # Certificate verification
├── integrations/    # PeerCert, GEPA, RLM, Mandala
├── messaging/       # MessageBox client, AuthSocket
├── server/          # Authenticated HTTP server
├── startup/         # First-run interactive setup
├── storage/         # Vaults, UHRP, memory manager
├── wallet/          # Identity Client adapter, PushDrop ops
├── types/           # Shared TypeScript types
├── start.ts         # Gateway entry point
└── index.ts         # Public API exports

packages/
└── openclaw-plugin/ # @agid/openclaw-plugin — standalone OpenClaw package

Development

npm run build        # Compile TypeScript
npm run dev          # Watch mode
npm test             # Run tests (vitest)
npm run test:watch   # Watch mode tests
npm run lint         # ESLint
npm run format       # Prettier

Security Model

• Authentication: Mutual authentication for all HTTP endpoints
• Encryption: ECDH end-to-end encryption for all MessageBox communication
• Key Derivation: Per-counterparty hierarchical key derivation
• At-Rest Encryption: AES-256-GCM for local vault storage
• Audit: Cryptographic hash chains with on-chain Merkle root anchoring
• Certificate Enforcement: Optional mode to reject uncertified senders
• No External Wallet API: All wallet operations are in-process only

License

Open BSV License

---

Built by BINARY.