Overview
What are Pods?
Pods are collaborative workspaces inside Dust that bring conversations, files, and tasks together around a shared goal or initiative. Everything your team and agents produce in a Pod stays in one place — searchable, organised, and ready to use as context for future work.
Built for Multiplayer AI
Most AI tools are built for a single person working with a single agent. That architecture breaks down quickly when work spans days, teams, and decisions that require more than one person to make.
Pods are built around a different premise: meaningful work is collaborative, and AI should be too.
Three kinds of collaboration happen inside a Pod:
- Human + human — teammates share conversations, files, and tasks. Everyone works from the same context, with no information trapped in someone's private chat history.
- Human + agent — you assign a task, an agent starts working on it, you review and steer. The agent has access to everything the Pod knows; you stay in the loop without losing the thread.
- Agent + agent — agents can hand work off to other agents, each picking up where the last left off, using the Pod's shared context as the connective tissue.
As the scope of what agents can handle grows — from minutes to hours to the equivalent of days of work — the bottleneck shifts from capability to coordination. Pods are designed to solve that coordination problem.
Guiding principles
Full transparency within a Pod
Everything inside a Pod is visible to all its members — conversations, files, linked data, tasks. There are no private corners within a Pod.
This is by design. Shared context only works if it is truly shared. When an agent runs inside a Pod, it can see what every member can see. When a teammate joins a Pod, they can read the full history from day one. No one is ever working from an incomplete picture.
If some work is sensitive and should not be shared with the whole Pod, it belongs in a separate Pod with the right membership — not hidden within an existing one.
Bi-directionality: everything a human can do, an agent can do
In a Pod, agents are first-class participants — not just tools you invoke. An agent can create a conversation, post a message, create a task, start working on it, mark it done, and upload files. The same actions available to a human member are available to an agent.
This means Pods can run with varying degrees of human involvement: from fully human-driven with agents assisting, to mostly automated with humans reviewing and steering at key moments.
The Pods skill
Agent interactions with Pods are powered by the Pods skill — a set of tools that lets agents read and write to a Pod's conversations, files, tasks, and metadata.
The Pods skill is available to all agents by default. This means agents can interact with Pods whether they are invoked from inside a Pod, from the global Dust agent, or from another workflow such as a trigger, scheduled run, or external integration.
One of the Pods skill's tools lets an agent list all Pods the user can access, making it possible to build cross-Pod workflows without hardcoding specific Pod references.
Capabilities
| Capability | What it enables |
|---|---|
| Group conversations by topic | Keep all conversations about a project or initiative together, out of the noise of your main feed |
| Give agents shared context | Upload files or link company data — agents in the Pod automatically use it when answering questions |
| Track and delegate tasks | Create tasks, assign them to teammates or agents, and track progress in one place |
| Control who has access | Invite specific members or make a Pod open to your entire organization |
Pods vs Connections: choosing the right data layer
Dust offers two ways to make knowledge available to agents: Connections (live-synced integrations with tools like Notion, Google Drive, and Confluence) and Pods (collaborative workspaces with a shared Files tab). Both support semantic search by agents, but they are designed for different situations.
| Connections | Pods | |
|---|---|---|
| Best for | Live-synced third-party knowledge (Notion, Google Drive, Confluence…) | Agent-first knowledge bases you control directly |
| How data gets in | Synced automatically from the connected tool | Uploaded manually, written by agents, or pushed via API |
| File format | Depends on the connected tool | Any format — including plain markdown files |
| Scale | Depends on connector limits | Scales to thousands of files; semantic search unaffected |
| Update model | Sync cadence driven by the connector | Immediate — files are available to agents as soon as they are written |
| Who manages the write pipeline | The third-party tool and its sync connector | You (or your agents) |
Use Pods when:
- Your knowledge base is a set of files — markdown, text, or other formats — that you push programmatically via the API, a scheduled agent, or an automated pipeline.
- You want agents to be first-class writers as well as readers: storing reports, summaries, or generated documents directly into the knowledge base.
- You need immediate availability — files written to a Pod are searchable by agents right away, with no sync delay.
Use Connections when:
- Your canonical knowledge already lives in Notion, Google Drive, Confluence, or another supported integration, and you want it kept in sync automatically.
- Your team edits documents in those tools and expects changes to be reflected in Dust without manual steps.