Manage multi-agent, multi-project work around tasks, not agents, with handoff
Problem
Multiple coding agents each keep their own work logs, so when one agent takes over from another, context must be re-explained. When juggling several projects, it is hard to know which tasks are done, in progress, or need human intervention without manually syncing across sessions.
Solution
A local-first task board where each task binds its context (goal, constraints, progress, artifacts). When a session ends, export a context package; a new agent imports it and continues. A board aggregates agent status and supports manual task state and intervention flags.
Always-visible quota and approval requests for AI coding agents on the Mac notch or edge
Problem
When running Claude Code, Codex, and other coding agents in parallel, you have to switch windows constantly to see which quota is running out and which agent is waiting for your approval. Developers report this breaks flow and that quota warnings come too late, causing rate-limit surprises at critical moments.
Solution
A tiny always-on Mac menu bar/notch app that aggregates quota usage, reset times, burn rate, and balance across agents, and surfaces pending approval requests with the command and file diff so you can approve or deny in one click. It alerts before quota runs out based on burn-rate projection.
Interactive audio ink-wash map explaining geography and history in ancient Chinese poetry
Problem
Parents teaching children ancient poems find that kids have no concept of place names or poets' journeys, as traditional text or static maps fail to engage. One developer spent a month making an audio ink-wash map, but ordinary parents cannot replicate that effort.
Solution
An online tool that selects common ancient poems and automatically generates an interactive map with audio narration. Users pick a poem, and the map dynamically shows the poet's footsteps and related locations, with ink-wash style visuals and a timeline. Audio explains background stories and geographic changes.
One-click export and migration of AI chat histories to prevent platform lock-in
Problem
A user lost nearly a year of chat history and research data when Zhipu's overseas site suddenly blocked Chinese users. Similarly, other platforms may change terms or impose regional restrictions anytime, and users often lack local backups, causing critical data to vanish. Users feel as distressed as losing a phone.
Solution
A desktop tool that exports conversation history from multiple AI chat platforms (ChatGPT, Claude, Gemini, Zhipu, etc.) into a universal format (JSON/Markdown) and supports importing into other platforms or local knowledge bases. It also offers scheduled automatic backups to ensure a local copy always exists.
LLM API capacity status dashboard with alerts and fallback suggestions
Problem
Developers using services like Codex often encounter 'Selected model is at capacity' errors and must manually try other models. Users complain this disrupts workflow, especially for paid accounts, and there's no easy way to check model availability or auto-switch.
Solution
A status monitoring service that probes major LLM APIs (OpenAI, Anthropic, Google, DeepSeek, etc.) in real-time for capacity, via simulated requests or public status pages, and presents an availability dashboard. Developers can subscribe to specific models, get notified when capacity is low, and receive fallback suggestions.
Lightweight self-hosted backup verifier that detects missing critical data
Problem
A developer lost code from his self-hosted Gitea after a disk failure. He thought his workspace and repositories were on different disks, but they were on the same one. Such incidents are common among self-hosters who manage backups manually and often miss key directories or wrongly assume coverage. Existing backup tools are either heavy or lack visualization and verification.
Solution
A self-hostable lightweight service that uses a simple config to declare directories and databases to back up. It periodically checks whether these are actually written to backup targets (another disk, NAS, cloud storage) and generates a visual coverage report. If a directory is missing or inconsistent, it sends immediate alerts.
Audit coding agent instruction files for consistency to prevent behavior drift
Problem
Developers find that as projects grow, global prompts and instruction files (AGENTS.md, .cursorrules, etc.) for AI coding agents become bloated and self-contradictory. For example, one rule says 'always format after coding' while another forbids modifications before commit. This leads to unpredictable agent behavior and degrading code quality. Users complain that global prompts have limited effect and cannot enforce compliance even after repeated trimming.
Solution
A local CLI tool that scans agent instruction files (including nested ones), parses rules, detects logical contradictions, duplications, outdated or conflicting requirements, and generates a readable report. It also tracks changes over time to alert on rule drift, helping developers maintain a clear, consistent behavioral contract for agents.
All your AI subscriptions' quotas and resets in one view
Problem
People juggling Codex, Gemini, Claude, and more lose track of how much quota is left and when it resets, forcing them to log into each dashboard separately and often missing a reset window. Cross-platform, cross-account quota tracking is entirely memory-based.
Solution
A local-first dashboard that aggregates authorized AI providers' balances and plan quotas, shows remaining usage and reset countdowns, and pushes alerts when a reset is near or a balance is low, with desktop and mobile widgets.
See only the branch you care about, no more rainbow commit spaghetti
Problem
With SourceGit or Fork open on multiple branches, the graph floods with colored vertical lines and interleaved commits, making it impossible to focus on one branch's history; users even suspect they are using it wrong. Managing all local repos and seeing multi-branch diffs in one place remains unsatisfied.
Solution
A desktop Git client that defaults to a "branch focus" view: selecting one branch shows only its linear commit list, collapsing other branches into a sidebar. When comparison is needed, explicitly overlay one or two branches with side-by-side or three-way diffs instead of permanent noise.
Get a phone notification when a page changes, stop staring at it
Problem
When waiting for scores, restocks, or ticket status updates, users install an auto-refresh extension and keep switching back to check. Refreshing is easy; knowing whether something changed still requires human eyes, and phones get no alert.
Solution
A browser extension refreshes pages on a schedule and watches for a specified text appearing/disappearing or a selected region changing. On trigger, it pushes a notification to the phone via Bark, ntfy, Telegram, WeCom bots, etc., and shows before/after content diffs.
Turn scattered lab reports into one clean trend line
Problem
People pile up yearly health-check reports, but when they want to see whether a value has been rising, different hospitals use different test names and units, making comparison impossible. Family members' reports and home glucose/uric-acid meter readings are also tracked by hand.
Solution
Scan or import PDF lab reports and auto-extract hospital, date, test name, result, unit, and reference range. Normalize tests with different names or units across hospitals into one trend line, pin out-of-range values to the top, and support separate profiles per family member plus recheck reminders and widgets.
See exactly which apps and sites eat your proxy traffic
Problem
Proxy subscription quotas quietly vanish every month. Clash Verge's connections page only shows currently open connections, keeps no history, and offers no daily or per-app breakdown, so users can only guess which app or domain burned the traffic.
Solution
A local daemon polls the Clash kernel every second, aggregates connection snapshots into SQLite, and reports traffic by app (merging child processes like Chrome Helper), by root domain, and by node/subscription, separating proxied from direct traffic with daily/weekly comparisons.
Catch AI-generated frontend bugs that never throw an error
Problem
AI-built stores and pages keep shipping with same-color text and backgrounds, client-side race conditions, and issues tests never exercise—usually discovered only after something breaks in production.
Solution
A tiny in-browser model plus scanner UI that checks generated pages for static and light runtime issues: contrast, style conflicts, client-side races—things that never throw—with copy-paste fixes.
One command to wipe stale branches, worktrees, and leftovers after a PR merge
Problem
Coding agents keep piling up stale worktrees and leftover branches until disks fill up. After every merged PR you must switch back to the default branch, delete branches, clean worktrees, then reinstall and rebuild—many steps, easy to forget.
Solution
A small CLI that remembers each project's cleanup flow and, after a PR merge, handles it all in one command: delete merged branches, prune worktrees, return to the default branch, and run project-specific clean, install, build, and run commands.
Let AI edit the document you already have open, no upload/download
Problem
Using AI on documents and spreadsheets usually means uploading the file, having the model read it in the background, and downloading a regenerated copy. The original file is never touched, and complex formatting, merged cells, and formulas often break in the rewrite.
Solution
An MCP bridge that connects AI clients to your open local document window so the AI edits in place, sharing the same UI, preserving original formatting and formulas.
Model your real delivery costs so AI services don't run at a loss
Problem
Founders selling per-call or per-minute AI receptionists and support agents have to reverse-engineer voice, carrier, and model costs from real call data to land at prices small businesses accept, like $49–199. Today that math is all manual.
Solution
A small cost-modeling tool: enter voice platform, carrier, and model rates plus typical call length, and it computes per-minute cost, break-even usage per plan, and a suggested price range.
Search, compare, and revisit every coding agent session from one local app
Problem
Developers using Claude Code, Cursor, Codex, and Copilot side by side have their sessions scattered across separate directories. Finding 'how did the agent change this last week' means opening each tool one by one, with no way to search or compare across them.
Solution
A local macOS app that reads session logs from all major coding agents, groups them by project, and supports full-text search, cross-session comparison, and export of key conversations. Everything stays on-device.
Reads local coding-agent sessions and converts user corrections into reusable skills so agents stop repeating mistakes.
Problem
Developers use multiple coding agents like Claude Code, Codex, and Cursor, but corrections to AI mistakes stay in the current session and agents repeat the same errors next time. Existing tools either don't read local sessions or require manual tidying, lacking automation. Users also demand strong privacy and consent controls.
Solution
SkillHarvest reads existing local coding-agent session logs, automatically identifies where users corrected the agent, and extracts structured skills (e.g., 'this project doesn't use any type'), which agents load next time via a skill mechanism. Every capture source is off by default with its own consent setting.
Diagnose and accelerate desktop software update checks (including macOS) so 'check for updates' no longer stalls for 10 seconds.
Problem
Users complain that many desktop apps and OS update checks take over 10 seconds, far longer than a network round-trip plus version comparison. Developers don't know where the bottleneck is, and users just tolerate the lag. Existing tools offer no observability into the update-check process.
Solution
FixMyMac is a menu bar utility that monitors each app's update-check requests, shows DNS, TLS, and HTTP phase timings, and identifies slow or timing-out update servers. For configurable apps, it offers options to skip redundant checks or use faster mirrors, plus a leaderboard of 'which app slows you down most'.
A Mac budgeting app that runs AI models on-device to answer spending questions without sending data to the cloud.
Problem
Existing budgeting apps either require an account and upload financial data to the cloud, or lack natural-language querying. Users want a fully local, no-account expense manager that sends no analytics or crash logs, but such products are rare—especially with local AI Q&A.
Solution
LocalBudget is a macOS (Apple Silicon) budgeting app where all AI models run locally; the ledger and questions never leave the device. Users can ask in natural language 'how much did I spend on coffee last month', and the app answers offline with no network requests.