Overview

Team chat with AI agents, apps, flows, and real-time sync

start.chat is a hosted communication platform where chat, AI, and automation live in one place. Servers, channels, and threads handle your conversations. Apps, flows, and agents handle everything else.

The source repository and self-hosting release are not public yet. See Software Availability and Rights for the current status.

Start here

Use this section for product concepts and how the pieces fit together. If you already know what you want to build, jump straight to the section that matches your goal:

Apps if you want to build integrations, UI, or backends inside start.chat
Flows if you want automations and scheduled work
AI if you want to configure HUD or external agents
Data if you want structured app storage and queries
SDK, CLI, or API if you want programmatic access from outside the app

Why start.chat

Programmable from the ground up

Open to build on through a full REST API, TypeScript SDK, CLI, and a rich app platform with UI blocks, backends, structured data, and OAuth. Build internal tools, dashboards, and integrations that live inside your chat.

Effective Agents

We don’t force our built-in AI on you, we work the agents you prefer.

For example, our OpenClaw plugin is powred by our first-class API/CLI/SDK with webhook and websocket support. Or easily spin up custom agents with start.chat, with unique profiles and per-channel configuration.

The system agent, HUD, includes a rich suite of abilities: vector search of messages, querying data, running app actions, using skills, and more.

Local-first

With our desktop app, your server can sync configuration, apps, flows, and more locally to your disk. Combine with our first class CLI, API, and SDK, plus agent skills, and you and any local agents are able to leverage and bridge everything two ways from your desktop to your teams workspace.

Multi-platform

Web, desktop, and mobile from one codebase. Desktop via Tauri (macOS, Windows, Linux) with local workspace sync. Mobile via React Native (iOS, Android).

Core model

Servers are the top-level unit — for teams, communities, or 1:1 conversations. Every user gets a personal server on signup.

Channels organize conversations within a server. Public, private, read-only, solo, shared between servers, or auto-threaded for forum-style discussions. Messages support markdown, attachments, reactions, mentions, link previews, and inline app content.

Threads branch off any message for focused discussion. AI-generated titles, reply counts, and unread indicators.

Roles control access at every level — who can see a channel, who can moderate, who can install apps. Admin and Team roles by default, custom roles when you need them.

Search works across all messages you can access, combining full-text, fuzzy, and semantic search for results that actually find what you’re looking for.

Extensions and automation

Apps connect external services. Built-in apps for GitHub, Linear, Stripe, Discord, Postmark, Supabase, and more — install with one click. Build custom apps with OAuth, webhooks, backends, and structured data.

Agents are AI personas in your workspace. HUD is built in — mention @hud anywhere, use the HUD channel, or open the sidebar. Configure per-channel with custom models, abilities, and system prompts. Add multiple agents to a channel, each with independent triggers. External agents connect via OpenClaw.

The API, SDK, and CLI expose every action programmatically. API keys inherit permissions from a role, keeping access control consistent whether an action comes from the UI, an agent, or a script.

Build on top

Apps aren’t just integrations — they’re the extension primitive. Sync data, provide actions, run scheduled backends, and render blocks (inline UI) directly in chat. Write blocks in a conversation, have HUD generate them, or edit locally via desktop sync.

Data gives each app structured key-value storage with configurable visibility — public to server members, private to the app, or secret. Apps use it to store synced records, dashboards use it to display metrics, and AI uses it for context.

Flows chain app actions into multi-step automations written in Pipedown, a plain-language DSL. Triggered by events, schedules, or manual commands. AI at each step. Human checkpoints via /wait. No drag-and-drop — just write what you want to happen.