@pons/module-agent

Agent runtime module for the Pons platform.

Overview

The agent module provides the core AI agent loop — receiving messages, assembling context, calling LLM providers, executing tools, and streaming responses back. It manages:

• Agent CRUD — create, update, delete agent configurations with persona definitions
• Session management — persistent conversation sessions with transcript history
• Agent loop — think-act-observe cycle with streaming output
• Skills — loadable skill packs with tools, system prompts, and permission scopes
• Interactions — mid-run approval requests and user questions
• Hooks — lifecycle hooks for agent events
• Long-term memory — workspace-scoped persistent memory for agents
• Permissions — scoped permission system for tool access

Prerequisites

• Deno v2.0+

Installation

deno install

Usage

As a Pons module (kernel-managed)

The agent module starts automatically when spawned by the kernel. Configure agents in your workspace:

workspace/
  agents/
    my-agent/
      agent.yaml     # Agent config (name, model, skills, etc.)
      soul.md        # Persona / system prompt
      agent.md       # Additional agent instructions

Standalone

deno run -A runner.ts

IPC Interface

Subscribes to

Topic	Description
inbound:message	Incoming user messages from any channel
session:close	Session termination signal
llm:stream:chunk	Streaming LLM response chunks
run:cancel	Cancel active agent run
interaction:resolved	User response to interaction request

Publishes to

Topic	Description
outbound:{channelType}	Agent responses, tool events, stream deltas

RPC Services

• agentManager — agent CRUD, config resolution
• sessionManager — session CRUD, transcripts

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT