@agenthifive/agenthifive

OpenClaw plugin for AgentHiFive vault integration. Gives AI agents secure, policy-governed access to user accounts through vault-managed credentials and a brokered API proxy.

What It Does

• 7 agent tools for executing API calls, managing connections, approvals, and downloads through the vault
• Native channel plugins for Telegram and Slack
• Step-up approval flow for sensitive actions (user approves via dashboard)
• Prompt injection with chunked API reference docs for connected services
• ES256 JWT auth with automatic background token refresh
• Setup wizard (ah5-setup) for bootstrapping agent auth

Quick Start

1. Install the plugin

openclaw plugins install @agenthifive/[email protected]

2. Install the setup CLI

npm install -g @agenthifive/[email protected]

3. Run setup

Run the setup wizard to bootstrap agent auth, discover connected services, write config, and apply the temporary credential proxying patch used until native OpenClaw support lands:

ah5-setup

Or non-interactively:

ah5-setup --base-url https://app.agenthifive.com --bootstrap-secret ah5b_...

The wizard automatically:

• Writes canonical AgentHiFive config to ~/.openclaw/openclaw.json
• Applies the temporary credential proxying patch to your OpenClaw installation

After setup:

• Changing models from the OpenClaw TUI /models picker is fine
• Avoid re-running OpenClaw onboard or models auth for vault-managed providers, because those flows expect local provider keys/tokens

Additional flags:

• --config-path <path> — explicit config file path
• --openclaw-dir <path> — explicit OpenClaw installation directory
• --mode remove — remove channels.agenthifive cleanly, then uninstall the plugin

4. Connect Services

Add OAuth connections (Google, Microsoft, Slack, Telegram, etc.) and configure policies in the AgentHiFive dashboard.

Tools

| Tool | Description | |------|-------------| | vault_execute | Execute an HTTP request through the vault proxy (Model B) | | request_permission | Request permission for a sensitive action | | request_capability | Request access to a new service/capability | | vault_await_approval | Wait for a step-up approval to be resolved | | vault_connections_list | List available connections and their status | | vault_connection_revoke | Revoke a connection immediately | | vault_download | Download binary files through the vault to disk |

Configuration

Canonical config lives under channels.agenthifive in your openclaw.json.

Minimal example:

{
  "channels": {
    "agenthifive": {
      "accounts": {
        "default": {
          "enabled": true,
          "baseUrl": "https://app.agenthifive.com",
          "auth": {
            "mode": "agent",
            "agentId": "agent_...",
            "privateKey": "base64-encoded-jwk"
          },
          "providers": {
            "telegram": {
              "enabled": true,
              "dmPolicy": "balanced",
              "allowFrom": []
            },
            "slack": {
              "enabled": true
            }
          }
        }
      }
    }
  },
  "plugins": {
    "enabled": true,
    "allow": ["agenthifive"],
    "entries": {
      "agenthifive": {
        "enabled": true
      }
    }
  }
}

AgentHiFive's generic plugin layer derives its runtime config from that channel section, so plugins.entries.agenthifive remains intentionally minimal.

Important keys under channels.agenthifive.accounts.<id>:

| Key | Type | Description | |-----|------|-------------| | baseUrl | string | AgentHiFive API base URL | | auth.mode | "agent" | "bearer" | Authentication mode | | auth.agentId | string | Agent ID (agent mode) | | auth.privateKey | string | Base64-encoded ES256 JWK (agent mode) | | auth.token | string | Bearer token (bearer mode) | | auth.tokenAudience | string | Token audience override (optional) | | pollTimeoutMs | number | Approval poll timeout (default: 300000) | | pollIntervalMs | number | Approval poll interval (default: 3000) | | providers.telegram.* | object | Telegram channel settings | | providers.slack.* | object | Slack channel settings |

LLM Credential Proxying

The setup wizard automatically patches OpenClaw's resolveApiKeyForProvider() to route LLM API calls through the vault, so agents don't need local API keys.

This patch-based flow is the officially supported path today for vault-managed LLM calls. It is a temporary compatibility layer while the upstream OpenClaw contribution for native support is being processed.

The patch is applied during ah5-setup and:

• Works with both npm installs (dist/ chunks) and source installs (src/)
• Creates a .bak backup of the patched file
• Is idempotent (safe to run multiple times)
• Uses the shared globalThis.__ah5_runtime bridge, so it is a no-op when the plugin is not installed

To re-apply manually (for example after an OpenClaw upgrade), re-run the setup wizard or see patches/README.md.

Removing AgentHiFive

Use the setup CLI rather than uninstalling the plugin first:

ah5-setup --mode remove

That removes channels.agenthifive from config and then uninstalls the plugin cleanly.

Programmatic Usage

The package also exports classes for use outside the plugin system:

import { VaultClient, VaultTokenManager, VaultActionProxy } from "@agenthifive/agenthifive";

// Direct API client
const client = new VaultClient({
  baseUrl: "https://app.agenthifive.com",
  auth: { mode: "bearer", token: "ah5t_..." },
});

// ES256 JWT token management
const tokenManager = new VaultTokenManager({
  baseUrl: "https://app.agenthifive.com",
  agentId: "agent_...",
  privateKey: jwk,
});
await tokenManager.init();

License

MIT