explaingit

cynchro/deepseekcli

13PythonAudience · vibe coderComplexity · 3/5ActiveSetup · moderate

TLDR

Python CLI named deep that turns a plain-language description into a working project using the DeepSeek API, with planning, generation, evaluation, and a phone-friendly web UI.

Mindmap

mindmap
  root((deep))
    Inputs
      Project description
      Task text file
      DeepSeek API key
    Outputs
      Generated project files
      Evaluation report
      Saved experience cache
    Use Cases
      Generate a project from a prompt
      Update an existing project
      Chat with the model
      Use from a phone via PWA
    Tech Stack
      Python
      DeepSeek API
      Tailscale
      trustme

Things people build with this

USE CASE 1

Generate a small FastAPI or Flask project from a one-line description using deep build

USE CASE 2

Update an existing project in place with deep update to add features like JWT auth

USE CASE 3

Run deep serve over Tailscale and install the web UI as a PWA on a phone

USE CASE 4

Hold a multi-turn chat with the DeepSeek model using deep ask in the REPL

Tech stack

PythonDeepSeekTailscalePWA

Getting it running

Difficulty · moderate Time to first run · 30min

You need a DeepSeek API key, and the HTTPS PWA flow needs the trustme package plus manual certificate trust on each phone.

In plain English

This repository, written in Python, holds a small command-line tool called deep from Cynchro Labs. It uses the DeepSeek API to turn a plain-language description into a working project. You describe what you want, and the tool plans the architecture, writes the code, saves the files, evaluates the result, and stores what it learned for next time. The documentation is in Spanish. You can install it with pip install deepseek-builder, or with shell installers on Linux and macOS, or with a PowerShell script on Windows. The first time you run deep, it asks for a DeepSeek API key and saves it for later. A DEEPSEEK_API_KEY environment variable can override the saved key. The main command is deep build, which takes either a quoted description or a -t text file. It runs five phases: planning, generation, file writing, evaluation, and learning. If evaluation fails, you can pass -f to make the tool retry the fix automatically. A --model flag lets you pick between the default model and deepseek-reasoner. Other commands extend the same idea. ask opens a chat with the model that keeps context across turns and compacts itself when it nears the model's limit. update edits an existing project in place. fix repairs the current project using the metadata stored in a .deep/ folder. show prints the saved task, plan, files, and evaluation. doctor checks the local environment. A serve command starts a web interface for using deep from a phone, with optional HTTPS via the trustme package so the page can be installed as a progressive web app. The README walks through pairing the desktop and the phone over Tailscale, plus the manual steps for trusting the self-signed certificate on Android and iOS.

Copy-paste prompts

Prompt 1
Install deepseek-builder from pip, set my DEEPSEEK_API_KEY, and run deep build for a FastAPI JWT API
Prompt 2
Use deep build with the deepseek-reasoner model and the -f flag to auto-fix failed evaluations
Prompt 3
Set up deep serve with HTTPS and the trustme cert so I can install it as a PWA on my Android phone
Prompt 4
Walk me through pairing my laptop and phone over Tailscale so deep serve is reachable on mobile data
Open on GitHub → Explain another repo

Generated 2026-05-22 · Model: sonnet-4-6 · Verify against the repo before relying on details.