What it is
Published on Y Combinator’s software GitHub organization under the MIT license. Y Combinator is a startup investor; there is no paid product behind it.
A team deploys it on its own cloud account and connects it to Slack. Each employee has a private space and each channel or project a shared one, with separate memory, files and credentials, so one team’s work does not leak into another’s. The same core can be driven by different assistant engines, including Claude Code and Codex.
It runs scheduled jobs, watches for changes, accepts triggers from other systems, and lets users build small internal web tools in chat and publish them to chosen colleagues. Credentials for connected tools are visible only in the space where they were granted.
What marketing teams should take from it
- It shows that a shared, controlled environment for assistants exists off the shelf, with shared written methods, access limited to each job, and a record of what ran.
- It contains no company content. The claims, the proof and the reviewers are still yours to supply.
- It needs an engineer to deploy and maintain, plus your own model provider keys and hosting. Marketing would be a user, not the installer.
- Support is what an open-source project offers, a security document and a community, with no vendor on the phone.
Where to start
Use it as a checklist rather than a purchase. Ask your company’s AI lead whether marketing’s assistants have shared written methods, access limited to each job and a record of what ran. If not, that list is the requirement, whether IT runs this or buys the same from a vendor.
If IT does run it, start with one channel, for one campaign, where the assistant holds the brief, assets and approvals for that campaign only, and a weekly scheduled job that assembles the account digest for the demand team.
Keep in mind
Infrastructure, not a marketing tool. The cost is engineering time and model usage, and the security review is IT’s job.