Take any of these five app ideas. They're for the trees.

tahir — Sun, 19 Apr 2026 06:28:58 +0000

I have five app ideas in my notebook for the trees. The trees shouldn't wait on me.

Here they are. Take any.

Build them better than I would have. CC0, no credit needed.

A note before the list. Almost every "tree planting" app is the same map: pins, counters, a feel-good number going up. The interesting work is somewhere else. In survival rates. In transparency. In showing people what they actually have around them and what they are losing. The five sketches below try to push into that gap.

If something like one of these already exists, drop the link in the comments, that's the best possible outcome. The goal here isn't novelty. The goal is open, citizen-facing, forkable tools that anyone can pick up and run. If a closed-source or paywalled version exists, an open alternative is still worth having.

1. Empty Lot Matchmaking
2. AI Tree Doctor
3. Neighborhood Carbon Map
4. Dead Tree Ledger
5. Street View Tree Census

1. Empty Lot Matchmaking — Tinder for would-be planters

One side: people with empty yards, vacant lots, apartment frontage.
Other side: volunteers who want to plant, organizations that donate saplings.

The app matches all three by location. On top of the match, a vision-language layer: based on the lot's climate zone, soil type, sun exposure, and water access, it tells you "these three species will survive here, these two will die without irrigation."

Why it matters

A Smithsonian-led meta-analysis of 176 tropical restoration sites found that on average only half of replanted trees survive five years, with mortality reaching 80% at some sites.

Mostly because of the romantic "anything green" attitude that puts the wrong species in the wrong place. The "how many trees we planted" number is already a lie. The real question is "how many lived." This app inverts that statistic. It turns the planting ritual into something close to science.

2. AI Tree Doctor — A plant pathologist in your pocket

A user registers a sapling they planted in the app. Every time they walk past it, they take a photo. A vision model reads health from the photo: yellowing leaves, insect damage, water stress, bark wounds. Each photo produces a care card: "water within three days, bark damage on the south side, check it."

Why it matters

A planted sapling's survival in the first years depends heavily on continued care, not just on planting quality. Fund4Trees' 2025 evaluation of urban planting in England explicitly calls for ending the "plant and walk away" culture.

People don't come back because they don't know plant pathology and have no feedback loop with that tree. Photo by photo, this app creates a "your tree is talking to you" moment.

3. Neighborhood Carbon Map — The street that carries you

A user draws a 500-meter circle around their home. They mark existing trees one by one, or auto-count them from street imagery (see #5). The app calculates the annual CO2 absorption of those trees and compares it to the user's own carbon footprint.

Why it matters

A mature oak captures around 25 kg of CO2 per year on average (the range is 10-40 kg depending on species and age); a person's annual flight footprint runs into thousands.

Someone who can see how much their street carries them starts thinking differently about urban planning, road widening, and cutting permits. This is not a tree map; it's a "right to the city" tool.

4. Dead Tree Ledger — A morgue for what we cut

Tree planting events get press. Tree death tracking happens almost nowhere. This app is a ledger for every tree that gets cut, dies of disease, or is destroyed by accident. The user uploads a photo, marks the location, and writes the cause if known.

The system auto-fills the local municipality contact form: "a tree was cut on this street on this date, when will the replacement be announced?"

Why it matters

Most municipalities cut tens of thousands of trees a year, and this data is centralized nowhere.

Planting is PR. Cutting is a quiet bureaucratic event. This asymmetry is what makes long-term tree loss invisible. An open, crowdsourced, photo-backed death registry becomes serious pressure on elected officials.

5. Street View Tree Census — AI excavating the past

A user enters a street or coordinate. The app pulls historical street imagery snapshots (2008, 2012, 2018, 2024, etc.) and runs them through a vision model. The model counts trees per year and produces a loss map: "this street lost 42% of its trees in fifteen years. Here is exactly where they were."

Why it matters

Academic teams have started doing satellite-based versions of this—like the Purdue group published in CACM in April 2026 on locating 278 million urban trees.

But none of this is at street level or in a citizen's hand. If a model can walk every street of a city in a few hours and produce a fifteen-year loss map that anyone can pull up by typing their address, that map gives activists and journalists a weapon they didn't have before.

A closing note

These shouldn't sit in a private notebook while trees are being cut down. If you are a developer with a free weekend and a working knowledge of any multimodal model, please pick one and ship it. Do it better than I would have.

If you start building one of these, drop your open-source repo link in the comments. Someone reading will want to collaborate, and this thread can be the meeting point.

In wealthier countries, some of this is partly under control. In much of the world, it isn't. There are people there who feel each cut as their own, organizing through small grassroots networks. If one of these tools lowers their friction even a little, that alone is reason enough.

The earth is our vein. The trees are our blood.

CC0. No credit needed. Just build.

Start an Open Source Repo for one of these ideas!

I replaced myself with DeepSeek as tech lead for Claude Code

tahir — Wed, 15 Apr 2026 17:30:27 +0000

Claude Code can write code better than most of us. But it still needs someone to lead — decide what to build next, break the task down, review the output, keep things moving.

I was spending more time managing Claude Code than it spent doing the actual work. Even with plan mode.

I was still the bottleneck.

So I built supervis.

What supervis does

supervis puts DeepSeek between you and Claude Code as a senior tech lead. You describe what you want once. DeepSeek plans the approach, delegates tasks to Claude Code, reviews results, and keeps going until the whole thing is done.

No researching which framework to use, no figuring out project structure, no learning state management patterns first. DeepSeek handles those decisions — you just tell it what you want built.

Live demo

I gave it one prompt:

"Have Claude build a markdown note-taking app with live preview and deploy it to Cloudflare Pages."

DeepSeek planned the architecture, told Claude to scaffold the project, build the editor, wire up live preview, handle persistence, run the build, and deploy. I didn't touch the keyboard once after the initial prompt.

The result:

🌐 Live app
📦 Source code — every commit from Claude Code, directed by DeepSeek

How it works

DeepSeek tells Claude Code what to do. Claude Code explores the codebase, writes code, runs builds, and reports back. DeepSeek reviews the results and decides the next step. You watch everything happen in real time through the TUI.

Under the hood, every task is a call to:

claude -p <prompt> --output-format stream-json --permission-mode bypassPermissions

Full transparency, zero interruptions.

Quick start

pipx install supervis
cd your-project
supervis

You need:

Claude Code installed (subscription is enough)
A DeepSeek API key

What's in v1.1.1

Live streaming display for DeepSeek output
Input history (up/down arrows)
Message queue management (/queue, /cancel)
Toggle reasoning display (/reasoning) to see DeepSeek think
File-based logging with --debug flag
Three model profiles: deepseek-chat with thinking, without thinking, and deepseek-reasoner

Open source

MIT licensed. Currently DeepSeek-only. If anyone wants to add support for other providers, PRs are welcome.

👉 github.com/arikusi/supervis

Forem: tahir

Take any of these five app ideas. They're for the trees.

Table Of Contents

1. Empty Lot Matchmaking — Tinder for would-be planters

Why it matters

2. AI Tree Doctor — A plant pathologist in your pocket

Why it matters

3. Neighborhood Carbon Map — The street that carries you

Why it matters

4. Dead Tree Ledger — A morgue for what we cut

Why it matters

5. Street View Tree Census — AI excavating the past

Why it matters

A closing note

I replaced myself with DeepSeek as tech lead for Claude Code

I was still the bottleneck.

What supervis does

Live demo

How it works

Quick start

What's in v1.1.1

Open source