Open sourceSelf-hosted. Bring your own model.

A crew of small bots that work on your machine

Give each bot one job. It works through the steps with your model, on its own computer, asks when a decision is yours, and remembers what it learns.
$curl -fsSL https://url.moduletrace.com/cobble | sudo bash
On a Linux machine with Docker. The installer adds Docker if it's missing.
Under the hood

Built from the harness up

cobble doesn't wrap someone else's agent. Every layer, from the loop that runs a bot to who may open which computer, is ours.

Our own harness

Written from scratch, not a wrapper around someone else's agent. Less overhead on every step.
Learn more

A model for every job

Text, media reading, transcription, speech and images, set per bot. Anything unset uses the defaults.
Learn more

Sessions, not one endless chat

Each topic gets a clean start that carries a summary of the last. Old sessions are searchable.
Learn more

Computers have owners

A bot reaches a computer through its workspace or a direct grant, and a new bot reaches none.
Learn more

Memory in layers

What holds everywhere, for you, and in this workspace. Only the layers that apply are loaded.
Learn more

Roles that contain the work

Members reach only the workspaces, bots and computers they're given.
Learn more
See every feature
Why cobble

The bots are the feature. Openness is the reason to switch.

Hosted bots run on someone else's computer, with someone else's model, and keep what they learn. cobble makes the opposite trade.

It runs where you do

Self-hosted on your own machine or server, so your logins stay with you. It's an install, not a subscription to someone else's computer.

You bring the model

Use the API key you already pay for, or a model running on your own hardware. Cobble runs models; it does not rent you one.

You own the memory

What a bot learns lives in a SQLite database on your server. Back it up, download it, restore it, and take it with you.

It is open source

The harness, the scheduler, the memory and the tool layer are MIT licensed, yours to inspect and fork. Nothing about how a bot works is hidden.
Calm, not chatty

The face is the status

No badges, no spinners, no running chips. Glance at the crew and you know who is working, who is done, and who needs you.
IdleReady for a job
WorkingTaking the steps
Needs youA decision is yours
DoneFinished, and says so
AsleepBack on schedule

Meet your first bot

Add a bot, give it a job in plain language, and come back to finished work. On your machine, with your key.