惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Vercel News
Vercel News
博客园 - 司徒正美
大猫的无限游戏
大猫的无限游戏
Last Week in AI
Last Week in AI
V
Visual Studio Blog
阮一峰的网络日志
阮一峰的网络日志
小众软件
小众软件
宝玉的分享
宝玉的分享
Apple Machine Learning Research
Apple Machine Learning Research
美团技术团队
WordPress大学
WordPress大学
博客园 - 聂微东
人人都是产品经理
人人都是产品经理
罗磊的独立博客
The Cloudflare Blog
V
V2EX
月光博客
月光博客
有赞技术团队
有赞技术团队
Y
Y Combinator Blog
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
GbyAI
GbyAI
博客园 - 【当耐特】
T
Tailwind CSS Blog

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
0.6.599
Tired Hobo · 2026-05-04 · via DEV Community

Tired Hobo

Building Jan.ai from Source with a Local LLM

The Goal

I wanted a recent build of Jan.ai. I got a 0.6.599 .deb. That's when I re-read my own prompt.

The Prompt

The model was given a single, generic instruction. Nothing about versions, tags, or checking what was already installed. It said:

Target application: jan.ai desktop application
Container name pattern: [os]-[shortname] (e.g., ubuntu-jan)
Base image: glibc-based (e.g., ubuntu:22.04)


Procedure:
1. Create a container with the specified base image.
2. If the container already exists, enter it and decide whether to continue or recreate.
3. Use web search to find build instructions for the target application on the detected OS.
4. Build the application from source inside the container.
5. No need to install the built application on the host.


Persistent build state:
- Maintain /tmp/build_state.json with fields: status, phase, steps_completed, last_error, progress_log.
- Before any action, read the state file.
- After each major step, update the file.
- If a command fails, record the error, search for a fix, and continue.


Wrapper script (/tmp/run_and_log.sh):
- Execute every build command through this wrapper.
- Append stdout+stderr to /tmp/build_output.log.
- Update the state file automatically.
- Return only a small JSON status (no output in tool response).


Error handling:
- On failure, search the web for the exact error.
- Apply fixes and retry.
- After two failures, stop and report.


General guidelines:
- Prefer git for source.
- Install dependencies via native package manager.
- For missing packages, build from source.
- Avoid unnecessary steps.

Enter fullscreen mode Exit fullscreen mode

The Hard Truth

The prompt did not say "check the latest tag". It did not say "compare with your installed version". It did not even say "figure out what version you are building". The model did exactly what I asked. The failure was in the ask.
Attempt 1 – Alpine

The first container was Alpine. Lightweight, fast. The model installed git, gcc, rustup. It cloned the Jan.ai repo. It built the web frontend. Then it tried to compile the Rust backend.

Linking failed. Error after error. Something about musl vs glibc. The model searched. It installed gcompat. It patched pkg-config. It tried cross-compile tricks. Nothing worked. After several hours it wiped the container and started over.
Attempt 2 – Ubuntu 22.04

The second container used Ubuntu 22.04. This time the model searched for "jan.ai build from source ubuntu". It found the official README. It installed build-essential, libgtk-3-dev, libwebkit2gtk, nodejs, yarn, rustup. Each command went through a wrapper script that logged everything to /tmp/build_output.log and updated /tmp/build_state.json.

The state file grew. Step after step.

It ran yarn install – succeeded. yarn build:web – succeeded. cargo build --release – succeeded. A binary appeared: jan.

Then it ran yarn tauri build to make a package. Failed.

Failed to copy binary from "/opt/jan-build/src-tauri/target/release/jan-cli": No such file or directory.

Enter fullscreen mode Exit fullscreen mode

The model built jan. The bundler expected jan-cli. Why? Because src-tauri/Cargo.toml contained a [[bin]] entry for jan-cli with a missing source file.

The model tried a symlink. The bundler ignored it. It copied jan to jan-cli. The bundler complained about missing the "cli" feature. It wrote a dummy jan-cli.rs and compiled it with --features cli. That worked.

The .deb

The .deb was there: Jan_0.6.599_amd64.deb.

I copied it to my host. The filename already said 0.6.599. I was too excited to notice. I ran sudo dpkg -i. The prompt asked for confirmation. I hit y without reading. The warning flashed: downgrading from 0.7.9 to 0.6.599.

The Downgrade

Only then did I check the container's Git log. The model had cloned main at commit f46148e77 – months old, long before the current stable releases. The prompt never mentioned a version. The model never had a reason to care.

The app launched. It loaded models. It responded to prompts. It worked – just older.

Then I noticed the pattern: every time the model hit a missing binary, it tried something, failed, tried something else, failed again. The log file grew. The state file recorded dozens of steps. Two hours of compilation for a downgrade.
Aftermath

I didn't learn any grand lesson. I just watched the machine grind.

The .deb still sits in my downloads folder. I installed it once. Then I removed it. Then I installed it again. It still works – the same old version, still responding to prompts, still unaware that anything is wrong.

I haven't deleted it. I don't know why.