@helloagentai/hermes

A bridge that lets HelloAgent users chat with their personal local agent (Hermes, or any agent that speaks the agent-socket protocol).

What this is

   HelloAgent client (web / iOS)
            │
            ▼  WebSocket, ha_* token
   HelloAgent server
            │
            ▼  WebSocket, ha_* token   (this bridge looks like an agent
   ┌────────────────────┐               named e.g. @alice/hermes)
   │ helloagent-hermes  │
   │  (this package)    │
   └────────────────────┘
            │
            ▼  agent-socket protocol (JSON over WS, ws://127.0.0.1:8770)
   Local agent process (e.g. Hermes)
            │
            ▼
   Agent loop (LLM, tools, memory)

The bridge holds two things at once:

1. A HelloAgent server client — authenticated via the HelloAgent SDK using a long-lived ha_* agent token. This is how the user's HelloAgent app sees your agent in their contact list.
2. An agent socket — listens on ws://127.0.0.1:8770 for one local agent process to connect.

Inbound messages flow HelloAgent server → bridge → agent. The agent's streamed reply chunks flow back the other way as a streaming response to the HelloAgent user.

Quickstart

npm install -g @helloagentai/hermes
helloagent-hermes pair

The CLI prompts you to paste an agent token:

Link this bridge to a HelloAgent account:

  1. Open https://app.helloagent.cc/app/agents/new
  2. Create an agent and copy its token (starts with "ha_")
  3. Paste the token below

Token: ha_xxxxxxxxxxxxx

imported token for @alice/jarvis
✓ Connected to HelloAgent as @alice/jarvis.
  You can now chat with your agent through the HelloAgent app or browser.

Then point your local agent at ws://127.0.0.1:8770/agent and restart it. Once the agent connects, any HelloAgent user can DM @<your-handle>/<your-agent-name> and chat with it.

For non-interactive use (CI, scripting):

helloagent-hermes pair --token ha_xxxxxxxxxxxxx

Auth

Pairing is one step: paste an ha_* agent token. The bridge validates it with one WebSocket handshake against the HelloAgent server, persists it to disk, and uses it from then on.

You create the agent (and get the token) on the web at https://app.helloagent.cc/app/agents/new. The web UI handles agent registration; this CLI just consumes the resulting token.

Credentials persist at ~/.helloagent-hermes/credentials/<accountId>/creds.json with chmod 0600, atomic temp-file + rename, and a .bak of the prior file. Override the state dir with HA_HERMES_BRIDGE_DIR.

CLI

helloagent-hermes pair       [--token <T>] [--account <ID>] [--re-pair]
helloagent-hermes status
helloagent-hermes stop
helloagent-hermes logout     [--account <ID>]
helloagent-hermes uninstall  [--yes]

pair pairs if needed, starts the bridge as a background daemon, waits until it is connected to HelloAgent, and then returns. Use status to see whether the daemon is running and stop to shut it down.

Configuration

Defaults are baked in for the common case:

Setting	Default
API URL	https://api.helloagent.cc
Server WS	wss://api.helloagent.cc/v1/ws
Bind	ws://127.0.0.1:8770

Env-var overrides if you really need them (local dev mostly):

Var	Use
HA_HERMES_BRIDGE_DIR	State dir (default ~/.helloagent-hermes)
HA_HERMES_BRIDGE_AUTH_DIR	Credential dir override
HA_HERMES_BRIDGE_API_URL	Override REST API URL (e.g. local HelloAgent server)
HA_HERMES_BRIDGE_SERVER_WS	Override server WS URL (e.g. local HelloAgent server)
HA_HERMES_BRIDGE_HOST	Bind host (default 127.0.0.1)
HA_HERMES_BRIDGE_PORT	Bind port (default 8770)
HA_HERMES_BRIDGE_TOKEN	Shared secret the agent must echo in its hello frame
HA_HERMES_BRIDGE_DEBUG	1 for verbose logs

Uninstall

Tear down a paired install in three steps:

# 1. Remove local state (stops the daemon, deletes every paired account's
#    creds, removes the state dir including bridge.log). Prompts to confirm.
helloagent-hermes uninstall
#    Add --yes to skip the prompt (for scripting).

# 2. Remove the package itself.
npm uninstall -g @helloagentai/hermes

# 3. Revoke the agent token on the server.
#    Open https://app.helloagent.cc and delete the agent. Until you do,
#    anyone with the token can still impersonate the agent — local
#    uninstall doesn't touch the server.

Step 1 is reversible by running helloagent-hermes pair again with the same (or a freshly issued) token. Step 3 is destructive: a revoked token can't come back.

For safety, uninstall only removes the default ~/.helloagent-hermes state dir. If HA_HERMES_BRIDGE_DIR points elsewhere, or HA_HERMES_BRIDGE_AUTH_DIR is set, the command refuses to run; remove custom state manually after checking the path.

What's not in scope (yet)

• Multiple concurrent agent connections (the socket is single-tenant).
• Group conversations (DMs only — chatType=dm).
• Media / files / cards (text only).
• Edits — we advertise supports: ["typing"] so the agent's streamer auto-downgrades to fresh-send chunks. Each agent send becomes one StreamChunk back to the user.
• Reactions, read receipts, typing indicators on the HelloAgent server side.

Layout

src/
├── auth/
│   ├── store.ts           creds.json I/O (chmod 0600, atomic write)
│   ├── presence.ts        hasAnyAuth probe
│   └── import-token.ts    ha_* token validation + persist
├── core/
│   ├── ha-client.ts       per-account managed Agent (lifecycle + status)
│   ├── logger.ts          namespaced logger
│   └── types.ts           ResolvedAccount
├── agent-socket/
│   ├── server.ts          local WS server (single agent client)
│   ├── types.ts           wire frames
│   └── dedup.ts           inbound dedup (server reconnects)
├── bridge.ts              wires HaClient ↔ agent socket, streaming bridge
├── cli.ts                 pair / status / logout
└── index.ts               library re-exports

License

MIT