BrainBeta

BrainBeta

Your self-hosted work assistant in the browser.

A private client for your team: chat on your host, call live work apps through MCP, and pick Command Code or your own model — including Azure AI Foundry. Not a public AI website, and not a RAG copy of Slack.

Open chatSee how it works

Example conversation: sprint blockers in ClickUp, a Slack nudge that waits for approval, then unread Gmail.

Works with

Official MCP connections — live tools, not a copied index. Set up once, then Connect from chat or Tools.

  • ClickUp
  • Slack
  • Asana
  • Gmail
  • Notion
  • Linear
  • Atlassian
  • Zernio
  • Sentry
  • dFlow
  • GitHub
  • Snowflake
  • MongoDB
  • MCP Toolbox
  • Rybbit
  • Bytebot

Architecture

Private client. Live tools. Your models.

The browser talks only to Brain. Brain stores its own product data, calls the model endpoint you choose, and reads or writes work systems through MCP — with approval on writes.

Brain logical architectureSigned-in browser talks to Brain on dFlow Enterprise. Brain stores product data in Postgres, calls Command Code or Azure AI Foundry for models, and fetches live work data through stateless MCP.PEOPLESigned-in staffBrowser / Chat / Tools / Models / SchedulesHTTPSAPPLICATION - BRAIN ON DFLOW ENTERPRISENext.jsUI on :3000eve agent runtimeSessions, streaming, MCP, approvalsBetter AuthWorkspaces, roles, HITLPRODUCT DATAINFERENCELIVE TOOLSPRODUCT DATABASEOperator PostgresAccounts, chats, playbooks,schedules - not a copy of Slackor SnowflakeMODEL PLANECommand CodeDefault BYOKAzure AI FoundryCustom / private BYOASTATELESS MCPLive fetch at request timeClickUp, Slack, AsanaGmail, GitHubSnowflake, MongoDBMCP Toolbox, dFlowRybbit, BytebotWrites require in-chat approvalBrain owns accounts and transcripts. Work data stays in the source systems. Models are called at an endpoint the organization chooses.BRAIN LOGICAL ARCHITECTURE - DFLOW ENTERPRISE - AUGUST 2026

  • 01

    Your Postgres

    Accounts, transcripts, playbooks, and schedules stay on the host. Brain does not copy Slack, ClickUp, or warehouse tables into a search index.

  • 02

    Models you choose

    Command Code built-ins, or any OpenAI-compatible endpoint — including Azure AI Foundry. Brain does not train models; fine-tuning stays at the model host.

  • 03

    Live MCP tools

    Tasks, mail, deploys, and SQL at request time. If a tool is disconnected, Brain cannot pretend it still has that data.

Read the architecture →

How it works

From empty host to useful chat.

Sign in, connect the apps you use, pick a model, and ask about the work in front of you. Brain reads live systems — it does not copy them into a search index — and asks for approval when something risky would run.

01

Chat that stays on your host

A browser workspace for talking to Brain — streaming replies, saved threads, and starters when you need a nudge.

  • Replies stream in as Brain works
  • History lives on this host — reopen, rename, search
  • Edit the last message or retry a failed turn

02

Connect the apps you already use

ClickUp, Slack, Asana, Gmail, Notion, Linear, Atlassian, Zernio, Sentry, dFlow, GitHub, Snowflake, MongoDB, MCP Toolbox, Rybbit, and Bytebot — set up once, sign in from the menu, then turn tools on when you need them.

  • See Connected, Connect, or Set up needed at a glance
  • Connect, Disconnect, or App settings without leaving chat
  • Browse loaded MCP tools after Connect

03

Pick the model for the moment

Switch among curated chat models or custom OpenAI-compatible endpoints (including Azure AI Foundry). Brain remembers your choice on this device.

  • Faster or stronger — choose per turn
  • Add Foundry, Ollama, or a company proxy on /models
  • Preference saved locally

04

Self-hosted on purpose

Run Brain on your machine or on Dokku / dFlow Enterprise. Models and sign-in stay on paths you control — not a public AI website and not a hosted gateway lock-in.

  • Command Code or a custom OpenAI-compatible endpoint
  • Better Auth sessions on your Postgres — no SQLite fallback
  • Built for a trusted local or team host

On this host

More than a chat window.

Brain also covers the everyday surfaces of a team assistant — spaces, tools, routines, approvals, and host policy.

  • 01

    Workspaces

    Personal and team spaces, invites, and an active workspace that scopes chats and MCP grants.

  • 02

    Tools catalog

    After you connect an MCP app, browse the tools it loaded — no chat turn required.

  • 03

    Playbooks & schedules

    Save reusable prompts and run morning briefs or scheduled playbooks on this host.

  • 04

    Approvals & Ask mode

    Mutating tools wait for in-chat approval. Switch to Ask and the assistant cannot reach Slack or Snowflake on that turn.

  • 05

    Instance controls

    Signup mode, licenses, and operator settings for who can join and how the host behaves.

Ready when you are.

Create the operator account on first boot, sign in, connect an app, and ask Brain about your work. Docs cover architecture, self-hosting, and environment setup.

Open chatOpen docs