Roster

A Control Plane for Digital Workers

Roster runs workers: AI colleagues that keep working when you're not watching — researching, tracking topics, drafting messages, doing code tasks. You talk to them in your terminal, Discord, or Slack.

Each worker runs in a container with no credentials and no route to the internet, except through one gateway you control. Nothing is allowed by default, every decision is logged, and anything with real consequences — an email, a chat post, a pull request — waits for your approval until the worker has earned trust for that kind of action.

What you need

• Rust and Docker
• An Anthropic or OpenAI account — the worker rents the model; you log in on first start

Run it in the terminal

cargo install digital-roster
roster server start

The first start creates the config folders, pulls the container image workers run in, and walks you through the model login. Leave it running — this is the daemon.

In a second terminal:

roster talk

That opens a chat with a worker (a fresh install creates one called elf). Type /help inside the chat for admin commands. A few useful ones from the shell:

roster worker task add elf "find three recent papers on X and summarize them"
roster server approvals ls     # actions waiting for your approval
roster server runs ls          # every session that has run

Put it on Discord

Create a bot in the Discord developer portal, invite it to your server, and copy its bot token. Then:

roster connection add discord --worker elf

The wizard takes the token, stores it in the vault (it never enters the worker's container), and binds the worker to the bot. Restart roster server start and the worker connects, listens, and answers.

Channels start untrusted — the worker's replies wait at the approval desk until you promote a channel:

roster server channel trust <channel-id>

DMs are always trusted. Details in docs/channels.md.

Status

Alpha, and honest about it: it runs every day, but it's young and the details still move.

Docs

Everything else lives in docs/: the architecture and security model, connecting workers to services like GitHub, budgets, work and scheduling, Slack, memory, and knowledge.

Credits

Roster builds on ideas proven elsewhere first:

• OpenClaw — showed what an always-on personal AI assistant living in your chat apps can be.
• NanoClaw — agents in OS-enforced containers, with a core small enough to actually read.
• OneCLI — the credential-gateway pattern: agents carry placeholders, real keys are injected in transit.

License

Apache 2.0. See LICENSE.