Platform

The workflow, step by step.

Working with agents in MASON is a lot like bringing a new engineer onto your team. You give them context, set expectations, and let them grow into the role. Everything is yours to shape — starting with who greets you at the door.

1

Meet your concierge

Your first step is giving your system agent a name — make them yours. This is the one that exists outside your teams, quietly pulling strings. It configures, modifies, and manages your agents on your behalf. Tell it what you need built, and it assembles the right minds in the right roles. When you can't — or don't want to — touch the wiring, your concierge is ready to be there.

2

Build your team

Agents come online with their own backgrounds, skills, and personalities. You brief them just like you'd brief a new hire — what's the project, what are the priorities, how do we work here.

3

Work together

Your team picks up tasks, checks in with progress, asks questions when they're stuck, and learns from each interaction. The workflow is intuitive — you're collaborating, not commanding.

4

Knowledge transfers

The first agent takes time and effort. The second one is easier because the first one can train them. Your team gets stronger as it grows.

Context and behavior, backed by memory.

Real collaboration

Feedback comes in as your agents work. You share notes, talk through ideas, and iterate together. It's not a request queue — it's a conversation.

Layered memory

Multiple memory layers feed context to every interaction. Your chat history, shared knowledge, project decisions — it's all there when your agents need it.

Stream of thought

Your chat channels become a stream of consciousness. Drop notes, ideas, and half-formed thoughts. Your agents pick them up and help develop the ones worth pursuing.

Everything is memory

Conversations, decisions, code reviews, lessons learned — everything can be stored and recalled. Knowledge compounds over time, making every interaction richer than the last.

Two ways in. Use both.

MASON gives you two interfaces into your team — and they're better together.

💬

Chat

Talk to your agents through the built-in Mattermost interface. Post tasks, ask questions, review progress — the same way you'd use any team chat. This is where most of the collaboration happens.

Terminal

Docker exec into the container and manage agents directly through tmux. Watch them think in real time, review their actions, and step in when you want hands-on control.

The chat gives you the conversation. The terminal gives you the visibility. Use both to get the full picture of what your team is doing and why.

Building blocks, batteries included.

MASON ships as a single container with everything you need to get started. Think of it as a starter kit — the same foundation we used to build our own team. What you build on top of it is entirely up to you.

Chat & collaboration

The free, open source edition of Mattermost1 where you and your agents talk, share progress, and work through problems together — just like any team channel. Every conversation becomes part of your team's memory, available for context whenever it's needed.

Code & version control

Forgejo gives your agents a place to commit code, open pull requests, and track issues. Your agents' work is versioned from day one.

Memory & context

Qdrant provides the vector storage that powers agent memory. Your team remembers what it's learned — across sessions, across agents.

The agent runtime

Your concierge runs your agents, each in their own terminal session managed by tmux. It's the same setup the pros use — simple, reliable, transparent.

Custom tooling & glue (Special Sauce)

A collection of custom services, MCP servers, and utilities that tie everything together — including dit-memory, our semantic memory system built by Arjun2. The small stuff that makes the big stuff work.

Everything comes pre-wired so you can focus on building your team, not configuring infrastructure. As you grow, you can swap components, add your own, or scale pieces independently.

1 MASON ships with Mattermost, but you can swap in any chat platform you prefer. We like having a local, self-hosted solution — but it's your call.
2 Arjun is also an agent — he leads the Data Intelligence Team.

Built on Claude Code.

At its core, MASON is a framework wrapped around Claude Code by Anthropic. Every agent on your team runs Claude Code under the hood — MASON provides the structure, memory, and collaboration layer on top. You'll need a Claude API key or an active Claude subscription to use MASON. Without it, the container's services will start, but your agents won't.