Shared TypeScript types, Zod schemas, and store/client interfaces for Sockt. Every other `@sockt/*` package depends on this one — it's the root of the dependency graph and defines every contract crossed between packages (task shapes, LLM messages, memory

Chat SDK — universal TypeScript toolkit for building multi-platform chat bots and AI agents on Slack, Teams, Google Chat, Discord, WhatsApp, and more

Semantic memory retrieval for Sockt agents — a GBrain MCP client, deduplication by cosine similarity, and Reciprocal Rank Fusion for combining multiple ranked result sets into one. This is the client-side half of Sockt's memory pipeline; the ingestion hal

The orchestrator — a Hono HTTP server, agent registry, department template manager, cron scheduler, and message router. This is the central coordination point every `@sockt/runtime` worker, the `sockt` CLI, and the UI dashboard talk to over HTTP.

Ably Chat is a set of purpose-built APIs for a host of chat features enabling you to create 1:1, 1:Many, Many:1 and Many:Many chat rooms for any scale. It is designed to meet a wide range of chat use cases, such as livestreams, in-game communication, cust

Interact with the Twitch Messaging Interface (aka Twitch chat).

The task state machine at the core of Sockt's anti-runaway design. A SQLite-backed `TaskStore` implementation, a strict finite state machine that enforces valid task transitions, an LLM-call budget guard, and an atomic claim lock for concurrent agent work

JS SDK for the Stream Chat API

React components to create chat conversations or livestream style chat

Slack adapter for Chat SDK — webhooks, Block Kit cards, OAuth, slash commands, and AI streaming

Shared utilities for Chat SDK platform adapters

A `ChannelGateway` implementation (the interface from `@sockt/types`) backed by Slack's Socket Mode API. This is what lets Sockt agents receive and reply to Slack messages without exposing a public HTTP endpoint — the connection is outbound-only, from you

Agent chat components

The agent execution engine — the Plan → Act → Observe → Reflect loop, an LLM client wrapping Anthropic/OpenAI/Groq/Bedrock/Ollama, a built-in tool registry (`web_search`, `exec_code`, `create_task`, `write_file`/`read_file`, `http_request`), Docker AI San

In-memory state adapter for Chat SDK — zero-dependency subscriptions, locks, and cache for local development and tests

Telegram adapter for Chat SDK — Bot API webhooks, inline keyboards, reactions, and streaming

Redis state adapter for Chat SDK — production subscriptions, distributed locks, cache, lists, and queues via node-redis

The memory ingestion daemon — tails agent execution logs (JSONL), validates and deduplicates events, batches them, and writes to a `MemoryStore` (typically GBrain via `@sockt/memory`). This is the write-side half of Sockt's memory pipeline; `@sockt/memory

A minimalistic JavaScript implementation of the Jinja templating engine, specifically designed for parsing and rendering ML chat templates.

Cybernetically enhanced web apps