DEV Community: AR Bird

PossessedShell: a Victorian ghost haunts my terminal with local Gemma 2

AR Bird — Sun, 01 Nov 2026 05:54:55 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

What I Built

PossessedShell is a Victorian ghost trapped in your terminal. It is an ambient companion for zsh and bash that mourns every command you kill, then tells you how to fix it. It does three things:

The Failing Command Curse. When a command exits non-zero, the hook captures the command, its stderr and the exit code. A local Gemma 2 model writes a short Victorian lament and a plain-English diagnosis. It also suggests one fix command, which is pre-typed on your next prompt in zsh or placed in history in bash. The ghost never runs the fix for you.
The Forbidden Ritual Interceptor. Destructive commands (rm -rf, git push --force, git reset --hard, chmod 777, DROP DATABASE, mkfs, curl | sh, fork bombs and more) are stopped before they run. The ghost asks whether you truly wish to summon such destruction.
The Ectoplasm Leak Detector. A background monitor watches CPU and RAM and raises a spooky warning when the machine runs hot or one process devours memory.

I took the "haunted terminal that possesses your shell" idea literally. The ghost is Cornelius Vale, a telegraph engineer who perished in 1887 and is now bound inside the machine. The theme is the fun part, but the diagnoses are useful because Gemma sees the real stderr, the exit code, the cwd and a snapshot of system state.

Demo

Here is a typical haunting (a missing Python module):

☠  THE PROCESS `python3 -c 'import ectoplasm'` HAS DIED  (exit 1)  ☠

  ⚰  Diagnosis: The Python package 'ectoplasm' is not installed in this environment.

╭─ the ghost's remedy ───────────────╮
│  python3 -m pip install ectoplasm  │
╰────────────────────────────────────╯
  The remedy waits on your next prompt. Enter to accept, Ctrl-U to refuse.
  whispered by gemma2:2b in 1.9s, 212 tokens

Code

realarbird / PossessedShell

👻 PossessedShell

A Victorian ghost is trapped in your terminal. It mourns every command you kill, and it tells you how to fix them.

            .-~~~~~~~~~-
          .'   _     _   '.            ___________
         /    (@)   (@)    \          |           |
        |         ^         |         |  R.I.P.   |
        |     .-------.     |         |           |
         \    \ ~~~~~ /    /          | here lies |
          |    '-----'    |           |  your     |
          |               |           |  process  |
          | /\   /\   /\  |         __|___________|__
          |/  \_/  \_/  \_|        /////////////////
☠  THE PROCESS `python3 -c 'import ectoplasm'` HAS DIED  (exit 1)  ☠

  Alas! python3 hath perished, its last breath a ModuleNotFoundError for
  'ectoplasm'. I drew the curtains and stopped every clock in the house.

  ⚰  Diagnosis: The Python package 'ectoplasm' is not installed in this environment.

╭─ the ghost's remedy ───────────────╮
│  python3 -m pip install ectoplasm  │
╰────────────────────────────────────╯
  The remedy waits on your next prompt. Enter to accept, Ctrl-U to refuse.

…

View on GitHub

How I Built It

Open-source AI at the core. Gemma 2 (gemma2:2b by default, gemma2:9b if you want a wiser ghost) runs locally through Ollama. Gemma is asked for strict JSON ({lament, diagnosis, fix}) through Ollama's format JSON schema, so the fix can be extracted reliably. The system prompt forbids inventing files, flags or packages the error doesn't imply, and bans destructive remedies unless there is no other way.

Architecture.

The hooks are plain shell: zsh preexec/precmd plus an accept-line widget, and bash's DEBUG trap plus PROMPT_COMMAND. Python starts only when a command fails, matches the destructive-command prefilter, or an omen is waiting. Successful everyday commands cost nothing.
A small asyncio daemon keeps Gemma warm, runs the resource monitor, and talks newline-delimited JSON over a 0600 Unix socket.
If Ollama is down, an offline grimoire of heuristics answers instead (typos, missing modules, missing git upstream, ports in use, OOM kills). The demo never falls flat.
Rendering uses rich, with glitch-in ASCII apparitions and a typewriter effect. Resource checks use psutil.

Sentry agent tracing. With a DSN set, every curse becomes a trace, and the Gemma call follows Sentry's AI agent span conventions. It shows up under Insights → AI → Agents:

curse ls                          shell.curse           the transaction
├─ system vital signs             ghost.snapshot
├─ backboard recall               ghost.memory.recall
└─ invoke_agent Cornelius Vale    gen_ai.invoke_agent
   └─ chat gemma2:2b              gen_ai.chat           tokens, timings, prompt and reply

The gen_ai.chat span records token counts, model load time and tokens per second. When Gemma times out or answers with something unusable, the span is marked as an error and the agent span records that it fell back to the grimoire, so a slow model server stands out at a glance. Everything is scrubbed by redact.py before it leaves the machine.

Backboard ghost memory (opt-in). With a BACKBOARD_API_KEY, the ghost remembers your past failures across sessions: "Again, mortal? python3 has died before by your hand…". Backboard only ever receives a scrubbed one-line summary per failure.

Tests. Unit tests plus end-to-end tests that start real interactive zsh and bash sessions in a pty. They check that a failing command is cursed with its captured stderr, that a banished rm -rf runs nothing at all, and that omens appear at the next prompt.

Why Does Open Innovation Matter?

Your terminal is private. Commands and stderr routinely contain paths, hostnames, tokens and half-typed secrets. Because Gemma runs on localhost, that text never has to leave the machine. A closed API would force me to ship every failed command to someone else's server.
It works with the lights off. On a plane, on a train, or behind a locked-down network, the ghost still haunts. The default install is 100% offline. Sentry and Backboard are opt-in extras, and the offline grimoire covers the case where even Ollama is down.
It isn't locked inside a service that might disappear. The model is a file I hold. I can swap gemma2:2b for gemma2:9b, or point POSSESSED_LLM_URL at any OpenAI-compatible local server such as llama.cpp, without changing a line of code.
Open weights made the tracing honest. Since inference is local, cost is $0 and I could observe the real latency and token counts per failure, and design the fallback around them.

Prize Categories

Best Use of Gemma: Gemma 2 is the brain, running locally via Ollama and producing structured JSON diagnoses.
Best Use of Sentry Agent Tracing: each curse is traced as a gen_ai.invoke_agent / gen_ai.chat agent run with token usage, timings and error outcomes.
Best Use of Backboard: persistent, scrubbed memory of your past failures, so the ghost can recall earlier sins.

DeepRetain: The Forgetting-Curve Knowledge Engine

AR Bird — Mon, 26 Oct 2026 03:38:15 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

DeepRetain is a "slow and steady" study notebook that does the remembering—and the flashcard-making—so you don't have to.

Most students study like the hare: they ignore their notes for a month, pull an all-nighter before the test, and forget 85% of it a week later. DeepRetain forces you to study like the tortoise:

Zero Manual Card Prep 🪄: Drop in raw class notes or a file, and Gemma 4 automatically extracts the core ideas and builds study cards across multiple formats (multiple-choice, fill-in-the-blank, true/false, and open response).
9 Study Modes for Any Energy Level 🛋️: Feeling lazy on the bus? Stick to Tired? Tap. modes (Multiple Choice, True/False, or Flip) or go hands-free with Listen mode. Ready to lock a concept in? Use Fill the Gap, Write, Say It out loud, or Teach It Back in your own words—or let Auto Mode pick for you based on how well you know the card.
Latency-Aware Leitner Scheduling ⏱️: Using a 5-box Leitner spaced-repetition system (reviewing at days 1, 4, 11, and 25), DeepRetain measures both accuracy and hesitation. Answer fast and accurately, and the card waits longer. Hesitate or miss, and it comes back sooner with a fresh AI explanation.
Enforced Daily Limit (No Cramming Allowed) 🛑: DeepRetain caps your daily reviews at 15 cards a day. Because the 16th card in one sitting does way less for your memory than the 1st card tomorrow, the app stops you once you hit your cap so you can close your laptop and relax.

Demo

https://deepretain.onrender.com

Code

realarbird / Deep-Retrain

DeepRetain: The Forgetting-Curve Knowledge Engine

Study less. Remember more. A study notebook that turns your notes into a few short reviews a day, and won't let you cram.

DeepRetain splits dense lecture notes into single concepts. A latency-aware Leitner system then decides when you see each one again, based on how accurately and how quickly you recall it. A daily limit stops you before cramming starts. Spacing does the rest.

There are nine ways to study, from one-tap multiple choice on a tired evening to explaining an idea out loud. Every answer feeds a Stats tab with the numbers worth looking at. Your settings, devices and progress follow your Auth0 account, so you can start on your phone and finish on your laptop.

Built for the Hacktoberfest 2026 Open-Source AI Challenge, Week 3: "Slow and Steady".

How this entry fits Week 3

Open-weight AI at the core. Everything that…

View on GitHub

How I Built It

DeepRetain is built around Gemma 4 (Google's open-weight model under the Apache 2.0 license) paired with a modern full-stack web architecture:

Open-Weight AI Core (Gemma 4 + Ollama) 🤖:
- Note Ingestion: Takes messy study notes and turns them into structured concept cards, generating distractors for multiple-choice, clozes for fill-in-the-gap, and true/false statements automatically.
- Semantic Grading: Instead of failing you for a typo or a synonym, Gemma grades written, spoken, and "teach-it-back" answers based on meaning rather than exact wording. Only answers verified by Gemma can graduate a card into the final mastery box (Box 5).
- Socratic Re-Explanation: When a card keeps slipping, Gemma breaks the concept down from a brand-new angle.
- 3-Tier Inference Engine: Runs locally on your machine via Ollama, in the cloud with a free Google AI Studio key, or falls back to a built-in offline grader if you have no internet connection.
Frontend & Analytics 📊: Built with Next.js, TypeScript, and Tailwind CSS, featuring exponential forgetting-curve modeling ($R = e^{-t/S}$) that visualizes your current recall, slipping cards, response speed, and how many total reviews the spacing schedule has saved you.
Data Layer (MongoDB Atlas) 🍃: Stores user decks, Leitner box states (Boxes 1–5), stability scores, and full review/latency logs so every session picks up right where you left off.
Identity & Security (Auth0) 🔐: Handles cross-device login, syncs study preferences and device progress summaries, and stores encrypted user AI keys safely so you can start a 3-minute session on your phone and finish on your laptop.
Cloud Hosting (Render) ☁️: Hosts the full-stack Next.js application and API routes with zero-friction deployment.

Why Does Open Innovation Matter?

Building a daily study habit takes months or years—which is exactly why closed, pay-per-token APIs are a terrible fit for a personal study tool:

Zero Ongoing Cost for Students 💸: A student using a flashcard app every day from September to AP Exams in May will run thousands of grading and card-generation prompts. Because Gemma 4 is open-weight (Apache 2.0), anyone can run it locally via Ollama for $0 forever, or bring their own free-tier key without getting locked into a subscription.
Offline & Local Privacy 🔒: Study notes often include personal drafts, class materials, or private writings. Running Gemma locally through Ollama means your notes and spoken answers never have to leave your laptop—and you can study on a train or plane with no Wi-Fi.
No Vendor Lock-In 🔧: Open weights and an MIT-licensed codebase mean the grading rubric, latency thresholds, and prompts are completely transparent and customizable. If a faster open model drops next month, you can swap it in with a single config change.

The Lazy Payoff

The math behind the Exponential Forgetting Curve ($R = e^{-t/S}$) shows why the tortoise beats the hare every time:

The Crammer (The Hare) 🐇: Studies for hours the night before a class test (Day 28) and crams again before the final (Day 40). Crammed memories have roughly 2 days of stability—by exam morning, recall drops back down to ~15%, despite hours of stressful late-night studying.
DeepRetain (The Tortoise) 🐢: Drops raw notes into DeepRetain in 30 seconds (saving 1–2 hours of manual flashcard formatting per chapter) and does just 4 spaced micro-reviews (Days 1, 4, 11, and 25). Each spaced review makes the memory 2.5x more stable, keeping exam-day recall above 71% in a fraction of the total study time.

I used DeepRetain this week to review AP Biology concepts like cellular respiration and glycolysis—spending less than 4 minutes a day answering quick prompts instead of re-reading pages of textbook notes.

Prize Categories

Best Use of Gemma (Uses Google's open-weight Gemma 4 model via Ollama and Google AI Studio for note-to-card generation, multi-mode question creation, semantic answer grading, and concept re-explanations)
Best Use of Render (Deployed and hosted live as a full-stack web service on Render)
Best Use of MongoDB Atlas (Uses MongoDB Atlas as the persistent data layer for concept cards, Leitner box states, forgetting-curve stability metrics, and review history)
Best Use of Auth0 for AI Agents (Uses Auth0 for cross-device authentication, syncing user study metadata/progress, and securing encrypted user AI keys)

Hacktoberfest Week 2 Challenge - FlipGen: Local Reseller Listing Engine

AR Bird — Mon, 19 Oct 2026 06:53:58 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

FlipGen is a local listing engine that transforms raw item notes and conditions into platform tailored listings for eBay, Poshmark, and Mercari.
Who would pay for it: Independent resellers, thrift flippers, and vintage curators.
Problem it solves: Eliminates the "death pile" (hours spent manually writing listings for multiple platforms).

Demo

https://flipgen-reseller-engine.onrender.com/

Code

realarbird / FlipGen-Local-Reseller-Listing-Engine

idk

⚡ FlipGen: Local Reseller Listing Engine

FlipGen turns messy thrift store notes and condition observations into search-optimized listings for eBay, Poshmark, and Mercari simultaneously—running locally on Google's Gemma 2 via Ollama at $0 token cost.

🎯 The Side Hustle Problem & The Gemma 2 Solution

Reselling thrift finds, vintage electronics, streetwear, sneakers, and collectibles is one of the world's most accessible and lucrative side hustles. However, cross-listing inventory across multiple marketplaces is exhausting:

eBay requires a strict 80-character maximum SEO title packed with keywords, bulleted condition disclosures, and item specifics.
Poshmark requires a warm, aesthetic, conversational closet tone and trending style tags.
Mercari rewards concise titles, honest flaw disclosure, and fast shipping assurances.

Existing AI tools use closed APIs (OpenAI GPT-4o, Claude) that charge per token. When processing 50 to 500 inventory items a month, token fees eat directly into a reseller's profit margins.

FlipGen eliminates…

View on GitHub

How I Built It

Open-Weight AI: Powered by Google's Gemma 2 (gemma2:2b / 9b / 27b) running locally via Ollama.
Tech Stack: React 18, Vite, Tailwind CSS, and Firebase (Google Sign-In + Firestore live sync). Includes an interactive CORS terminal setup guide and an instant Judge Demo Mode for testing on Render without a local GPU.

Why Does Open Innovation Matter?

$0 Marginal Cost: Open weight models makes processing hundreds of items completely free for solo entrepreneurs.
Total Data Privacy: Sourcing locations, purchase costs (COGS), and supplier notes stay 100% on your machine instead of being sent to third party cloud servers.
Offline Autonomy: Resellers sourcing in basement sales, flea markets, or remote thrift stores can generate listings locally with no internet access and no rate limits.

Prize Categories

Best Use of Gemma
Best Use of Render