Crewly

The problem

AI assistants live outside the place work actually happens.

For most teams, work happens in Slack threads. Decisions, context, follow-ups, code review chatter. But the AI tools those teams use live somewhere else: a browser tab, a separate app, a chat window with no memory of the team. Every request starts with copy-paste and re-explaining.

The result is that AI gets used for one-off questions, not for actual work. The assistant never learns how the team writes, what the customers prefer, or what was decided last week.

The approach

@mention. Crewly picks the skill. You get the work back.

Crewly installs into Slack and works inside the threads the team already uses. You @mention what you need. Crewly picks the right skill, asks if it needs more context, and gets to work.

The differentiator is memory. Crewly keeps an editable team memory: how the team writes, which clients prefer async, what tone fits which channel. Replies are source-cited, workspace context is permissioned, and customer data is never used for model training.

What we learned

Skills beat prompts. Memory beats context windows.

The early prototype was a general chat bot with a long system prompt. It demoed well and worked badly. The rebuild around discrete skills, each with its own context assembly and output contract, is what made results predictable enough to trust in a real workspace.

Team memory had to be editable to be trusted. When users can see and correct what the AI remembers, they stop treating its output as a black box and start treating it like a coworker's draft.

Result

Live, installable, three minutes to first useful output.

Crewly is live at crewly.live with a 14-day free trial, no credit card required. Install to Slack, @mention Crewly in a thread, and the first draft comes back with sources and the memory it used.

The skill roadmap ships in releases: four more specialists are queued, and MCP integrations connect Crewly to the tools the team already runs.

Stack

Technologies used