OpenClaw Integration

AgentHiFive integrates with OpenClaw to provide secure, auditable authority delegation for AI agents. Instead of storing OAuth refresh tokens on the agent host, AgentHiFive acts as the credential boundary: the agent runtime never sees provider secrets, and every action is policy-checked and audit-logged.

The problem today

OpenClaw agents commonly use external CLIs (like gog for Google Workspace) and store auth profiles locally. On VPS and remote gateway deployments this creates two issues:

• UX friction -- localhost OAuth callbacks require SSH tunnels or device-code workarounds.
• Security risk -- refresh tokens live "on the box," adjacent to a prompt-injectable model runtime.

AgentHiFive eliminates both problems. Provider tokens stay in the Vault; the agent host stores only short-lived session credentials and opaque connection IDs.

Three integration surfaces

AgentHiFive ships three complementary integration surfaces. You can adopt them progressively.

Surface A: Gateway Plugin (recommended)

The @agenthifive/openclaw npm package is a first-class OpenClaw Gateway plugin. It runs in-process with the Gateway (trusted boundary) and registers agenthifive.* tools the agent can call.

• Best UX and security posture.
• Bundled skill (SKILL.md) teaches the agent safe usage patterns automatically.
• Stores at most Vault session credentials and connection_id references -- no provider tokens.

See the Plugin Guide for installation and configuration.

Surface B: CLI Compatibility Layer

The agentgog CLI is a drop-in replacement for gog (the Google Workspace CLI) that delegates execution to AgentHiFive instead of storing tokens locally. This is the fastest way to upgrade existing OpenClaw workflows without rewriting skills.

• Implements the subset of gog commands that OpenClaw skills rely on.
• Local disk stores only connection_id mappings, not provider tokens.

See the Vault GOG Strategy for the implementation roadmap.

Surface C: MCP Server (portability)

The @agenthifive/mcp server exposes connect, execute, and revoke tools over the Model Context Protocol. This lets OpenClaw and other agent clients (Claude Code, OpenCode, etc.) consume AgentHiFive capabilities portably.

See the MCP Server guide for setup instructions.

Target users

User	Use case
OpenClaw operators (local or VPS)	Secure provider access without localhost callbacks or token-on-disk risk
Skill authors	Build tools against agenthifive.* instead of managing OAuth directly
Security-conscious end users	Enforce separation between model runtime and secrets

Key benefits

No tokens on the box

Provider refresh tokens are stored exclusively in the AgentHiFive Vault. The OpenClaw host never sees them, even momentarily. This holds true for local, VPS, and sandboxed deployments.

Policy engine

Every execution request is checked against configurable policies before reaching the provider API:

• URL allowlists and method restrictions
• Payload and response size caps
• Redaction filters for sensitive data (emails, attachments, secrets)
• Risk-based step-up approvals for destructive actions

Audit trail

Every action returns an audit_id that correlates across AgentHiFive and OpenClaw logs. This enables compliance reporting, incident investigation, and user-facing transparency.

Remote-friendly connect flow

Connecting a provider account works the same way on a laptop and a headless VPS:

1. User initiates connection via the AgentHiFive dashboard or API.
2. User opens the OAuth URL on any device (phone, laptop) and approves.
3. Agent can list connections via connections_list to discover available connection_ids.
4. No localhost callback, no SSH tunnel, no device-code workaround.

Core UX flows

Connect a provider account

User:  "Connect my Gmail"
User:  initiates connection via dashboard → approves OAuth
Agent: calls agenthifive.connections_list → discovers connection_id
Agent: "Connected: Gmail (read-only)"

Execute a read operation

Agent: calls agenthifive.execute with a GET request
Vault: checks allowlist, enforces caps, redacts sensitive fields
Agent: receives structured result + audit_id

Execute a write with approval ("high-five")

Agent: drafts the write action
Agent: calls vault_execute → vault returns 202 (approval required)
User:  approves via dashboard
Agent: re-calls vault_execute with approvalId → action executes
Agent: receives result + audit_id

Security model

Boundary	Trust level	What it stores
LLM / agent runtime	Untrusted (prompt-injection assumed)	Nothing sensitive
OpenClaw Gateway + plugin	Trusted (user-controlled host)	Vault session credentials, connection IDs
AgentHiFive Vault	Credential boundary	Provider refresh tokens, policies, audit log

Default safety posture

• Read-only by default.
• Step-up approval required for destructive actions (send, delete, modify, share, export).
• Generic HTTP calls restricted to allowlisted base URLs with strict caps and redaction.

What's next

• Plugin Guide -- install and configure the Gateway plugin
• MCP Server -- set up the MCP server for Claude Code and other clients
• Vault GOG Strategy -- roadmap for vault-backed gog CLI