Skip Ruby, Postgres, and Redis — a single-file SQLite forum that just runs
Problem
A V2EX developer said every attempt to give a project a Discourse-style forum died at the setup screen: Ruby + PostgreSQL + Redis, multi-gigabyte images. Small projects, communities, and internal teams need a serviceable forum but are forced onto heavy infrastructure.
Solution
A forum that starts from a single binary or small zip, defaults to SQLite, and mirrors Discourse's default theme in look and interaction: topic list, timeline, notifications, unread, badges, polls. Built-in signup/login plus OIDC/OAuth, server-side rendering, and SEO essentials like sitemap, RSS, and JSON-LD.
Auto-reconnects and monitors all your local SSH tunnels, alerting the moment one drops
Problem
A V2EX developer described needing to reach 4–5 production databases simultaneously: if even one SSH tunnel is forgotten or drops, the whole service errors out and they have to debug hop by hop, only to find the tunnel again. SSH tunnels work, but managing them is painful, and network blips require manual restarts.
Solution
A desktop or menu-barresident SSH tunnel manager: centrally configure all port forwards, auto-reconnect with exponential backoff on disconnect, show every tunnel's health on one screen, and fire a system notification or webhook when one drops. Optional local port-conflict detection and latency charting.
Routes Claude Code / Codex completion events to smart lights, menu bar, and phones
Problem
Multiple Show HN and Product Hunt launches all do the same thing: alert you when an agent finishes, via a smart light, a desktop character, or a menu bar icon. But each tool binds to one device or platform, so users install several just to get one notification. What developers actually want is a unified agent-completion event bus where they pick the outputs.
Solution
A lightweight local daemon watches Claude Code/Codex hooks or logs, normalizes events into done / needs-confirmation / error, and pushes them via webhook to any channel: smart lights, Mac menu bar, Telegram, Bark, and more. Users configure routing rules themselves without touching agent code.
Scans AI coding agent logs for leaked secrets and tells you exactly what to rotate
Problem
In the Warp security incident, users were told their agent traces might contain secret keys and to consider rotating them — but individuals had no way to know which keys leaked or what to rotate. Developers routinely feed entire codebases and terminal context to agents like Claude Code or Codex, and API keys and DB passwords easily end up in traces and logs.
Solution
Run regex and entropy scanning locally or in CI over agent session logs, trace files, and terminal transcripts to detect common API keys, tokens, and connection strings. On a hit, show exactly where it leaked, which service the key belongs to, and generate an actionable rotation checklist. Ship as a CLI plus a lightweight web panel with scheduled scans.
One gateway to securely expose all your devices' MCP capabilities to any AI client
Problem
Users have laptops, phones, and servers, each with a pile of MCP servers, skills, and plugins, but every AI client requires reconfiguring them separately. Switching from ChatGPT to Claude means doing it all over again, and there is no unified entry point or permission layer.
Solution
A self-hosted unified MCP gateway: a lightweight agent on each device registers with the gateway, which aggregates and exposes a single MCP endpoint. Users connect their AI client once and get permission-controlled access to command execution, file management, and browser automation across devices—no reconfiguration when switching clients.
Publish your indie project to multiple showcase sites with one submission
Problem
AI has multiplied the number of people building products, yet many side projects end up forgotten in a photo album or on GitHub. Some showcase sites exist, but they rely on users manually filling out forms and uploading screenshots again and again, making curation expensive.
Solution
A standardized project metadata format and submission API: developers maintain one project profile (name, link, screenshots, category) and push it to any connected showcase site via API or a static file, with changes synced automatically. Showcase sites get structured data and cover images without manual entry.
Daily visibility checks for your apps across App Stores, with instant takedown alerts
Problem
An indie developer discovered their app had been removed from the App Store only when they searched for it in the morning, with no heads-up beyond a backend notice. There is no lightweight tool that continuously answers “can people still find my app?” until users start complaining.
Solution
A lightweight monitoring service: users add an app's store ID and regions, the system periodically scrapes search visibility and detail page status from each regional App Store, and sends alerts via email, Telegram, or webhook when the app can't be found or the page returns an error. It saves page snapshots as evidence for appeals.
Send 10GB files directly between friends, bypassing cloud drive throttling and chat app limits
Problem
A user wanted to share about 10GB of music with a friend, but WeChat caps single-file size, Baidu Netdisk throttles non-VIP downloads to 20KB/s, and sharing via NAS is bottlenecked by the ISP's upload bandwidth. Existing options are either slow, size-limited, or require technical setup on both ends.
Solution
A zero-config desktop and browser P2P transfer tool: both sides open a link to establish an encrypted direct connection with automatic best-path selection. Large files are chunked and transferred in parallel with resumable support, and the receiver can download directly in a browser without installing a client.
Track wasted tokens from failed AI agent runs and auto-generate refund requests
Problem
Heavy users of Claude and other AI agents frequently see tasks fail mid-run with connection errors, yet their usage quota and tokens are still consumed. There is no resume button—only retry, which starts over and burns more tokens if it fails again. Users neither know how much they lost nor how to ask for compensation.
Solution
A local usage guardian that automatically attributes tokens consumed by failed agent tasks from API and proxy logs, produces a readable loss report, and generates ready-to-send support ticket templates. It can also halt further retries after repeated failures to prevent runaway spending.
A Bazi charting API that handles true solar time, DST, and solar-term boundaries
Problem
Users ask Claude or ChatGPT for Bazi charts, but AI often fails around solar-term transitions, birth-location true solar time, historical daylight saving, and early/late Zi-hour rules. Developers who build their own calendar hit the same edge cases.
Solution
A no-login remote MCP endpoint and API that correctly handles true solar time, historical DST, solar-term transitions, and Zi-hour rules, returning structured four-pillar charts with explanations for AI and developer integration.
A minimal personal site for writers who want their own distribution, not a feed
Problem
Writers want an independent site but worry nobody visits in the age of AI summaries and algorithmic feeds. Some build one and get single-digit reads, while domain, styling, RSS, and publishing overhead pile up.
Solution
A ready-to-run self-hosted publishing tool focused on writing and distribution: lightweight pages, RSS, email subscriptions, and cross-platform syndication, with no tracking or ads, pulling readers back to the author's own site.
Protect deep work from AI news and social feed rabbit holes
Problem
Developers get pulled away by AI news, GitHub hot takes, vulnerability alerts, and sales promos. One planned to learn TypeScript but got derailed by a new tool, a server sale, and a money-making side gig, losing days.
Solution
A browser extension plus focus panel that delays feeds by topic and urgency, sets a daily information budget, blocks non-urgent content during focus blocks, and only lets allowlisted sources through immediately.
Hard caps and real-time circuit breakers for every AI agent tool call and dollar
Problem
As AI agents automatically call paid APIs, external tools, and payment endpoints, developers fear runaway spend and unauthorized actions. Existing options cover model tokens or catch overruns after the fact.
Solution
A lightweight proxy layer that routes agent tool calls and API requests through a firewall, enforcing budgets, call limits, allowlists, and approval rules in real time, with logs and a spend dashboard.
Turn coding-agent effectiveness into a comparable score on real repo tasks
Problem
Developers want to know whether their coding agent is actually getting better, but rely on vibes or model switching. People clean repos, add lint rules, and feed better context without any way to measure if it helps.
Solution
A local benchmark tool that pulls verifiable micro-tasks from a real repository, runs them across different agents or configurations, and records pass rate, time, token usage, and human interventions into a comparison report.
Do small spreadsheet work in plain text that both AI and humans can edit
Problem
Small tasks like tracking a dog's health records or organizing tax documents feel overkill in a full spreadsheet app, while Markdown tables have no formulas. Worse, when AI is asked to help inside a spreadsheet it usually needs spreadsheet-specific tools or export formats, so it cannot directly read and edit the workbook.
Solution
Define a compact, readable plain-text spreadsheet format with basic formulas and cross-sheet references, plus a lightweight editor. Because the file is plain text, AI can read, modify and explain it directly, while humans get a grid view in the editor.
See exactly what your browser agent clicked and why it failed
Problem
When running browser agents, cost and speed are top pain points, and once a step misclicks or stalls it is hard to reproduce and diagnose. Most approaches rely on screenshots for the model to read, making element recognition flaky and debugging opaque.
Solution
A lightweight local debug panel that captures each step's element list, chosen action and latency, records failing steps with single-step replay, and lets developers compare how different models or strategies decide at the same step to find selector or prompt issues fast.
Spot and mute AI-generated cold outreach before it hits your phone
Problem
Founders receive a growing flood of clearly AI-generated, useless unsolicited emails that Google's spam filter does not flag, so their phones buzz all day, especially in the morning. Existing subscription filters are pricey or require handing mailbox access to a third party, and simple Google scripts are hard to maintain or extend.
Solution
A self-hostable mail filtering layer that connects via Gmail/IMAP, uses light models plus rules to recognize AI cold-email traits (template phrasing, bulk headers, no prior thread), auto-archives or mutes by sender domain, and sends a daily digest while real client mail still notifies normally.
A client-site delivery log and knowledge base for forward deployed engineers
Problem
The AI boom has created strong demand for forward deployed engineers who embed with clients, understand requirements and ship custom solutions. They gain unique visibility into client environments and problems, but those insights scatter across chats, meetings and personal notes, vanish when a project ends, and are hard to reuse or use to prove value; many also feel treated as glorified consultants.
Solution
A lightweight workbench for field delivery: log environment quirks, pitfalls, custom changes and decision rationale per client/project, auto-generate reusable delivery handbooks and weekly reports, and build a searchable private knowledge base for the next project or teammate.
One shortcut to turn rough thoughts into natural English in any input box
Problem
Non-native English speakers at global companies think in their native language but must type English in Slack, email and docs, re-reading messages without knowing how to fix awkward phrasing. Generic AI tools like Raycast have cramped input boxes, awkward prompt editing, and force window-switching and copy-pasting that breaks the workflow.
Solution
A lightweight Mac menu-bar app invoked by a global hotkey in any input field: feed it Chinese or rough English, get natural English, press enter and it fills back into the original box. Runs local models offline on Apple Silicon or cloud models like DeepSeek/Claude/OpenAI, with built-in presets for polish, rewrite and tone.
Highly accessible Hacker News reader for visually impaired and reading-difficulty users
Problem
Hacker News' simple UI is not friendly to visually impaired or dyslexic users: lacks screen reader optimization, font adjustments, high-contrast modes. Existing third-party clients rarely focus on accessibility.
Solution
An HN reader focused on visual accessibility: screen reader optimizations, adjustable font/line height, high-contrast themes, keyboard navigation, and clean mode. Available as a web app or browser extension.