cd /news/ai-agents/from-a-thought-in-bed-to-open-source… · home topics ai-agents article
[ARTICLE · art-98513] src=dev.to ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

From "a thought in bed" to open source: one week of development diary

A developer built DeepSeek Phone Harness, an open-source tool that lets users control a DeepSeek agent on their computer from a phone via HTTP and DSH RPC. The project, released under MIT, includes tool cards for step-by-step visibility, a remote terminal, file browsing, and task history, totaling about 4,000 lines with zero npm dependencies. The developer fixed a hanging issue caused by unanswered agent questions and emphasized the value of watching the agent work remotely.

read2 min views1 publishedAug 16, 2026

No glamorous start — just a late-night thought. A week of building DeepSeek Phone Harness, from zero to release.

In bed that night, I realized I could have the agent on my computer run a task early. Get up and walk to the study? Too much trouble. Give up? Unsatisfying.

In that moment it hit me: the agent's value is doing work for me, yet I'm chained to the computer.

Next morning I opened the editor and decided to solve it myself.

The core idea was simple: the phone sends HTTP, an agent process translates it into DSH RPC, and translates responses back.

That evening the first version worked: send a sentence from the phone, the agent on the computer executes, the result returns to the phone. Ugly, but it moved.

On day three, tasks started "randomly hanging." I dug until late night and found: the agent asks the user questions, questions don't go through the approval channel, and I wasn't listening to the question channel. Tasks waited forever on a question nobody answered.

When it was fixed, my commit message said three words: "no more hangs."

Results weren't enough — I wanted the phone to see what the agent does, step by step.

So tool cards were born: one card per tool call, with icon, status, the file path it touched, and expandable full arguments and results. The agent's reasoning also folds into a collapsible block.

That night I sat watching the agent work on my phone for a long time. "Watching it think and move" versus "waiting for a result" are two completely different experiences.

Added the terminal (remote commands), file browsing, task history (one-tap re-run), and session management (rename/delete/search).

Wrote the README, wrote the promo posts, pushed to GitHub. There was no ceremony — just git push

and a long exhale.

~4000 lines, zero npm dependencies, one Node process + one HTML file.

It runs stably: dispatch tasks on the commute, tweak config before bed, tap approvals mid-meeting. The phone became the agent's remote — "wherever you are, your agent is too" is no longer just a slogan.

Open source (MIT), three-minute setup:

git clone https://github.com/2903077918-lgtm/DeepSeek-phone-harness
cd DeepSeek-phone-harness && cp config.example.json config.json
node agent.mjs --mode=both

Install Tailscale on the phone, access over 4G/5G.

If you were to build an open-source project "for yourself," what pain point would you solve first? Comment below — the next viral project might start here.

── more in #ai-agents 4 stories · sorted by recency
── more on @deepseek 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/from-a-thought-in-be…] indexed:0 read:2min 2026-08-16 ·