Docs

Build plugins for AgentParley

Everything you need to write, publish and run a plugin: the SDK, the sandbox, every hook and provider, and real examples. And how to connect your own computers and local models through the tunnel.

Start

LLM start here

Plain-text versions of these docs for AI assistants and agents, and how to point yours at them.

Quickstart — your first plugin

Write a plugin, publish it, install it, and watch it change an agent's reply.

Plugin SDK

What a plugin is, what it can change, and how the pieces fit.

Concepts

Sandbox model

What your plugin can and cannot reach, and the memory, time and network limits it runs under.

Permissions

Permissions come from the code you wrote. How they're derived, what the installer sees, and how updates are gated.

Publishing

Drafts, immutable versions, private-first listing, yanking, and how installs follow updates.

Reference

Package format

The folder layout, readme.md, the entry file, TypeScript, and the size limits a package must meet.

Manifest reference

Every key of globalThis.manifest, what it does, and how it is validated.

Host API reference

Every member of the host object, its arguments, its result, and when it is present.

API reference

The REST endpoints for publishing and installing plugins, and the webhook address plugin channels receive on.

Guides

Tools

Give your agents a new action the model can call: a description, a JSON schema and a handler, running in the plugin sandbox.

Middleware hooks

The 17 points where a plugin can rewrite or veto what an agent is doing, as it happens.

Lifecycle events

Run code when your plugin is installed, updated or removed, and when a chat starts, is cleared or deleted.

Commands and skills

Ship slash commands and on-demand instructions as plain markdown files — no code.

Channels

Connect agents to a chat app — people message the agent there, and it answers there.

Live connections

Socket-only platforms without running a server — AgentParley holds the connection and wakes your plugin only when something matters.

Model providers

Make your vendor's chat or decision models selectable on AgentParley, like OpenAI or Anthropic.

Web search and web fetch providers

Replace the engine behind the built-in web_search and web_fetch tools.

Compaction provider

Replace how long conversations are summarized so long missions keep the thread.

Agent crews

Ship a team, not a tool — create and heal agents, shape how they delegate, and govern what they say to each other.

Create a 3D companion

For 3D artists. Put your character on screen in a few minutes, then share her with other people. No code needed.

Create a 2D companion

For Live2D artists. Put your rigged character on screen in a few minutes, then share her with other people. No code needed.

Companion packs

Ship 3D and 2.5D (Live2D) characters with their own motions, expressions, thinking sounds and behaviour. Data only, no code.

Voice and speech-to-text providers

Give companions a new voice vendor, or a new way to hear the user.

Examples

Example: Caveman

The smallest real plugin — one middleware hook that changes how every agent writes.

Example: Brave Search

A bring-your-own-key web search provider — settings, secrets, egress and host.fetch.

Example: Pi Compactor

A structured, iterative context summary written with the agent's own model — no key, no permissions.

Example: Discord

A full chat channel over a held socket — resumable Gateway, filters, heartbeats, idempotent sends.

Example: Oh My OpenAgents

A seven-agent crew in one package — provisioned, self-healing, model-matched, with doctrine, commands and skills.

Example: Sample Companion Pack

A data-only pack — an avatar look with its own motions, and two ready-made characters.

Example: OpenAI Voice

A voice and speech-to-text provider — binary audio in and out through host.fetch, typed vendor failures.

Running crews

Agent packs

Install a prebuilt crew — agents, doctrine, commands and skills — in one click.