Works where your team works
Each coworker is its own Slack app with its own name and avatar. Mention it in a channel or send it a direct message, and reply in the thread to keep the conversation going.
A coworker is a named Claude Code agent that your team hands work to. People @mention it in Slack, it runs jobs on a schedule, and you can give it a task from the console. It has a job description, it remembers what it learned between tasks, and every task runs in its own Droneship sandbox under your project’s egress policy.
@leo draft three subject lines for the webinar reminder
Leo reacts with 👀, answers in the thread, and keeps a note of anything worth remembering for next time.
Works where your team works
Each coworker is its own Slack app with its own name and avatar. Mention it in a channel or send it a direct message, and reply in the thread to keep the conversation going.
Runs jobs on a schedule
A weekly report, a morning pipeline digest, a daily KPI check. A schedule can post its result in a channel, or post only when there is something new.
Remembers between tasks
Its job description, progress and notes live in a GitHub repo you choose, so you can read and review everything it keeps.
Reaches only what you allow
Every task runs in a fresh sandbox. You pick its connectors and built-in tools, and the Highflame gateway enforces that list outside the sandbox.
| Sandbox | A fresh sandbox in this project for every task, with a fresh clone of its notes repo. |
| Runs as | Its creator: their GitHub connection, connectors and saved model key. Tasks count toward the project’s sandbox hours. |
| Model calls | Through the Highflame gateway. |
| Time limit | 20 minutes per task by default. Choose 10, 20, 30 or 60 minutes in Settings. A task that runs longer is stopped and marked failed. |
| Internet | Standard (the default) allows package registries and its tools. Strict allows only the AI model and its tools. Open allows the whole internet. The project’s egress rules also apply. |
| One task per thread | A thread runs one task at a time. Different threads run at the same time. |
Every run’s request and reply are visible to everyone in the project, including tasks that came from a direct message.