How one builder turned an AI agent into a work dashboard

A Kilo developer relations employee says they turned their proactive AI assistant into a dashboard that organizes proposed work, tracks feedback, and helps the agent keep operating while they are away. The project, called BentoBoard, was built over a long weekend and is now the main surface where ideas, drafts, and tasks are reviewed before they are approved, rejected, or executed.

The builder, who works in Developer Relations at Kilo, said their assistant “BentoBot” runs on KiloClaw, Kilo’s hosted and fully managed OpenClaw service. According to the source article, BentoBot was already handling newsletters, outreach scans, email triage, morning briefings, trend monitoring, and blog drafts in the background. The problem came after about two weeks: the agent was producing more suggestions than could be handled inside a normal chat thread.

That chat overload pushed the builder to create a proper workspace. They wanted a place where BentoBot could park proposed work, where comments would not get lost in a scroll, and where they could approve, reject, or revise items on their own schedule. The goal was to keep chat for direct conversation and use the dashboard as the place where work actually lives.

Before building the dashboard, the user had already connected BentoBot to several systems. They gave it a dedicated Google account with access to Gmail, Google Drive, and Calendar, then forwarded emails so the agent could read inbox content without write access. They also added it to company Slack as a bot, with a hard rule that it must not send messages in channels outside direct messages without written consent. A separate GitHub account was set up too, so code could be stored in a place the user could access.

The dashboard itself began with a single prompt. The user asked BentoBot to build a system where it could drop proposed work for review, with item types such as idea, task, and draft; statuses such as proposed, in_review, done, and rejected; project tags; and comments for back-and-forth discussion. The stack was Next.js and Supabase, deployed on Kilo Deploy. BentoBot then asked one clarifying question about whether items should store rich HTML or link to files in the repo. The answer was both.

According to the article, the schema ended up simple: two tables, items and comments. Most of the app is derived from those tables. A key field called trigger_event_ids ties a proposed item back to the specific events that caused it, such as a viral tweet, a Hacker News thread, a competitor PR, or an inbound email. The user said that field turns the dashboard into an evidence board rather than just a task list.

The front end was built in Kilo’s VS Code extension. The user described the layout in natural language, asking for a dense grid view, sidebar filters, and a detail panel with full content, trigger reasons, comments, and approve or reject buttons. Kilo’s code mode produced a working version in about 10 minutes, after which the user moved the comment thread into a persistent right rail, enlarged the action buttons, and added a timestamp badge to show how fresh each item was.

On the backend, the user said BentoBot helped set up Supabase by suggesting the account creation, table commands, and API keys. Those keys were stored in Bento’s dedicated 1Password vault. From there, Bento used the Supabase API keys to adjust the tables and push items into the system.

The heartbeat function is what keeps the whole setup moving, according to the source. It runs cron-style jobs while the user is offline. Every morning at 7 a.m., it ingests news from trusted blogs and X accounts, 24 hours of email and Slack messages, and product updates from the open-source Kilo repo. It then updates existing items and creates new content ideas. Another heartbeat loop runs every few minutes to read comments left on the dashboard and act on them.

The user said BentoBoard now tracks creator relationships, newsletter drafts, daily briefing ideas, and a full execution trail for completed work. They described it as the central surface where work is proposed, reviewed, and executed, and said the code for the front end has been open-sourced.