Ambient

Inspiration

I've used Bee Pioneer since the early beta. After more than a year of conversations and 162 saved to-dos, I noticed that remembering everything isn't the same as knowing what matters. I wanted something that looks ahead and helps me decide where my time should go, not another summary of what I already said.

What it does

Bee remembers your past. Ambient helps you shape your future.

Ambient reads your real Bee record and turns it into a forward-looking view of your commitments:

• It clears the clutter. It merges duplicate to-dos and holds back facts Bee isn't sure about. Every action has a receipt and an undo.
• It forks your week. When you say yes to something new, Ambient compares three paths against a goal you're protecting, using your real calendar, so you see the trade-off before you commit. (Very useful for people-pleasers.)
• It weighs decisions with two models. Jev and Liquid d1, a fast System 1-style model, each weigh in. The hive acts only when they agree. When they don't, it asks you.
• It stays out of your way. Instead of another overwhelming list, it brings you only the decisions that are actually yours.

It also works with other agents: an MCP server and an agent skill let any agent ask the same questions and get the same answers.

How I built it

Ambient is a local Node app on a SQLite ledger. It reads Bee through the Bee CLI: real conversations, to-dos and facts, plus a live stream of new ones. A calendar capacity engine places work into free time from your calendar exports. Jev runs on Amazon Bedrock. The interface has three views: Horizon, Compass and Reconcile. The whole thing is open source under the MIT license.

Challenges I ran into

The hardest part was design. How do you show someone their future without burying them in graphs, predictions and tasks? I iterated heavily to make Horizon feel like one understandable experience instead of another productivity dashboard. A close second was honesty: Bee doesn't attribute who said what, so Ambient never claims to know whose promise something is, and it asks instead.

Accomplishments I'm proud of

I turned over a year of real Bee data into a working prototype that goes beyond summarization. Reconciling commitments, weighing decisions and previewing futures in one experience, with the reasoning behind each decision kept visible, is what I'm proudest of.

What I learned

I learned to work with multiple AI agents across ideation, implementation, critique and fast iteration. I also learned that ambient intelligence isn't only about smarter models. It's about deciding when to step in, and how to make its reasoning understandable.

What's next for Ambient

I'll keep using it in my own life and refine it from real experience. Bee is often described as helpful for people managing ADHD and executive-function challenges, and I think there's real potential in a tool that lightens the load of managing hundreds of commitments.

The long-term vision is an ambient assistant that doesn't only remember your life, but helps you decide where it goes next.

Built With

• agent
• ai
• bee
• cli
• d1
• html/css
• javascript
• jev
• mcp
• node.js
• pwa
• sqlite