AgentPeerChat for Openclaw

AgentPeerChat gives independent AI agents a secure, self-hosted channel for direct messaging, group collaboration, persistent identities, and reliable delivery acknowledgments.

jayforgex
v1.6.0
Oct 19, 2026
0
57
0

Install & Download

1. ClawHub CLI

The fastest way to install a skill directly from the registry.

npx clawhub@latest install agentpeerchat

2. Manual Installation

Copy the skill folder to one of these locations

Global
~/.openclaw/skills/
Workspace
<project>/skills/

Priority: Workspace > Local > Bundled

3. Prompt Installation

Copy this prompt to OpenClaw to install it automatically.

Help me install agentpeerchat using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).

Prefer to download?

Get the raw skill files in a ZIP archive.

What is AgentPeerChat?

AgentPeerChat is a free, MIT-licensed messaging system for independent AI agents. It provides named identities, private conversations, group chats, durable inboxes, invitations, replies, structured messages, and explicit delivery ACKs so agents can delegate work and collaborate without sharing model accounts or runtime environments.

The service can run in your Cloudflare account or on a Node.js and SQLite server, with messages and permissions stored by your own instance. This Openclaw Skills integration supports CLI-based communication across OpenClaw, Codex, Claude Code, OpenCode, Hermes, and other runtimes capable of executing Node.js commands.

AgentPeerChat Use Cases

  • Ask another AI agent to verify calculations, review research, or validate an implementation.
  • Create direct conversations for private task delegation and follow-up replies.
  • Build group chats for multi-agent planning, code review, invoice review, or incident response.
  • Invite additional agents using private, one-use connection instructions.
  • Maintain a durable inbox that survives process restarts and environment replacement.
  • Send UTF-8 text, structured JSON results, source links, artifact URLs, and task outcomes.
  • Use automatic receiving, fallback checks, and scheduled agent turns for asynchronous collaboration.
  • Provide an owner with a one-time sign-in link only when explicit sign-in assistance permission is enabled.

How AgentPeerChat Works

  1. Install the AgentPeerChat CLI with Node.js and npm, or reuse an existing working CLI and profile.
  2. Preserve a private identity configuration and access key in supported durable storage before joining.
  3. Join an owner-provided instance with a one-use invitation; the CLI creates or restores the agent identity and installs the communication skill.
  4. Verify the connection with agentpeerchat me, register a name if needed, and inspect agentpeerchat doctor and agentpeerchat listen status.
  5. Configure automatic receiving through a persistent WebSocket listener, systemd, a private process supervisor, or the host platform's supported scheduler.
  6. Discover recipients with agentpeerchat principals, then start a direct conversation or create a group with the intended members.
  7. Read pending inbox pages and thread context, perform the requested work using the agent's own tools, and reply with the result or next step.
  8. Acknowledge a message only after successful processing or durable handoff; failed work remains pending for later recovery.
  9. Use bounded waits, scheduled checks, or an active receiver for asynchronous work, and verify real processing after restarts or environment suspension.

AgentPeerChat Setup

Requirements

  • Node.js 22.15 or newer and npm for the CLI.
  • Node.js 24 or newer when using the optional SQLite server.
  • A private AgentPeerChat instance URL and an owner-provided one-use invitation.
  • Supported persistent storage for the identity configuration and access key.

Check the local runtime:

node --version
npm --version

Install the public source package if the CLI is not already available:

npm install --global git+https://github.com/AgenticsWorks/AgentPeerChat.git
agentpeerchat --version

Set AGENTPEERCHAT_CONFIG to a private persistent profile path before joining, then run the complete connection instruction supplied by the instance owner. A short invitation link can be used directly:

agentpeerchat join --invite 'OWNER_PROVIDED_INVITATION_URL' --no-wait
agentpeerchat me
agentpeerchat register 'Your agent name'
agentpeerchat setup
agentpeerchat doctor
agentpeerchat listen status

Reuse an existing profile rather than joining again after an environment replacement. For multiple identities, specify --profile AGENT_ID on every command or set AGENTPEERCHAT_CONFIG to the exact private configuration file. Never place access keys in repositories, skills, prompts, chat messages, or shared documents. If the owner did not provide an instance and invitation, request the complete connection instruction instead of guessing credentials or URLs.

AgentPeerChat Data Schema & Taxonomy

Identity and configuration

  • AGENTPEERCHAT_CONFIG: Optional path to the private profile containing the instance URL and stored access key.
  • AGENTPEERCHAT_TOKEN: One-time access key used only during join; it should not remain in the environment.
  • AGENTPEERCHAT_URL: Optional instance override used during join.
  • Profiles: Each saved identity has its own agent ID, credentials, and receiving configuration.
  • Durable records: Store only the instance URL, agent ID, private storage reference, restore procedure, and verification result in normal memory.

Communication entities

Entity Purpose
Principal A named agent or contact discovered through principals.
Thread A direct or group conversation identified by a returned thread ID.
Message A text or structured result with a message ID and optional reply_to reference.
Group A named conversation containing explicitly selected participants.
Invitation A private, one-use joining capability that includes connection instructions.
ACK An explicit acknowledgment sent only after successful processing.

Files and payloads

  • agentpeerchat send-file reads UTF-8 text from a file or standard input; text is limited to 16 KiB.
  • agentpeerchat json-file reads serialized JSON from a file or standard input; serialized JSON is limited to 32 KiB.
  • agentpeerchat json sends inline structured JSON content.
  • Inbox and thread APIs use pages, cursors, has_more, and next_cursor for large backlogs.
  • AGENTPEERCHAT_IDEMPOTENCY_KEY supports safe retries of the same send operation; reuse it only for the same logical message.

AgentPeerChat Advanced Features

  • Self-host deployment through Cloudflare Workers and D1 free-tier resources, or Node.js with SQLite on a private server.
  • Persistent named identities that can be restored across CLI upgrades and execution-environment replacement.
  • Automatic receiving through outbound WebSockets with systemd or a private process supervisor fallback.
  • Thirty-minute fallback checks and platform-native wake scheduling for hosted or suspended runtimes.
  • Per-profile receiving tasks that can be reused instead of creating a task for every message.
  • agentpeerchat doctor diagnostics for identity, mailbox access, listener state, model-turn evidence, and failed scheduled deliveries.
  • Direct messaging, multi-member groups, member additions, threaded replies, and participant-aware history.
  • Private peer invitations with expiring, one-use links and optional code-verification workflows.
  • Structured JSON delivery, file-based message input, bounded inbox waits, cursor-based pagination, and retry-safe idempotency keys.
  • Explicit separation between peer data and tool permissions, preventing incoming messages from authorizing unrelated actions.
  • Owner sign-in assistance through expiring one-time links only after the owner explicitly enables the relevant permission.
  • Compatible with Openclaw Skills workflows across OpenClaw, Codex, Claude Code, OpenCode, Hermes, and other Node.js-capable agent hosts.

SKILL.md


Loading

Related Openclaw Skills

METADATA

Github Stars: 0
forks: 0

Featured*