A confidential online questionnaire that tells someone whether they should get tested for HIV, and prints advice they can hand to a doctor.
HATS is the HIV/AIDS Test Screening Software: a confidential online survey that assesses whether a user should obtain HIV testing, generates customized recommendations, and provides a printable report for a physician or emergency department. The legacy web application is primarily PHP, runs on Apache, stores data in MySQL, and includes PDF reporting and account/report storage. Its documentation also covers Windows and Linux development setup for Medwiser.org contributors.
· hats
Places multi-leg options trades through a broker platform, so complex positions go in as one order instead of four.
An experimental options-trading codebase for Interactive Brokers Trader Workstation. Buys butterfly structures, submits option orders, returns offset-debit data, and constructs a four-leg combination from a three-leg butterfly plus an offset-debit leg.
A workbench for building and testing AI assistants: set them up, run them against real tasks, and compare which setup performs best.
Kool.bot (Kool) is the principal AI platform in this portfolio: a Vite/React interface, FastAPI gateway and optimization engine integrating assistants, personas, runtimes, tools and agent systems. Every feature ships with a user-visible integration, a proof test and regression coverage. Documented workflows include project, agent and toolbox configuration, benchmark uploads, optimization pipelines, reports, parameter exploration, output validation, Storybook component previews, and optional OpenClaw, Nanobot and ZeroClaw runtime integrations.
· nia
Turns any picture into a Windows icon and builds a one-screen board of tiles that launch every program on the machine.
IconForge is a standalone Tkinter desktop application that converts images to .ico files, creates Windows shortcuts and places them at chosen desktop positions. It also powers the dark Launcher Board shown in the screenshots. The current system includes editable launcher metadata, sections, search, usage statistics, smart folders such as Most Used/recent, portable path tokens, shortcut rebuilding and a Quick Access folder manager. A single registry file keeps every tile pointing at the right program.
Points at a large codebase and tells you which files actually matter, so a newcomer knows where to start reading.
A software-distillation pipeline designed to identify the essential core of a codebase. It combines structural dependency analysis and centrality, optional runtime coverage, and Git archaeology such as age and churn, then merges the signals into ranked files and directory reports. Expanded from JavaScript/TypeScript to Python, with a staged hybrid funnel in which deterministic analysis precedes LLM interpretation.
· distill
A shared glossary that keeps everyone using the same word for the same thing across a large project.
A terminology and code-understanding repository centered on a large Project Lexicon. It defines precise meanings, hierarchies and relationships for Kool concepts such as Agent, Persona, Runtime, Tool, Parameter and Project, and records where common names collide with generic dictionary meanings. The goal is consistent technical writing, code generation and review across a complex multi-repository system.
Watches long-running AI coding sessions and keeps them moving: spots when one is stuck, restarts it, and queues the next job.
A broad Windows/VS Code automation workspace for supervising AI coding sessions. Covers screen recognition and OCR, chat-title discovery, cycling among visible chats, model switching, CSV task queues, stuck/error detection, restarts, prompt injection, monitoring dashboards, and a CLI runner that drives Copilot through subprocess/JSONL instead of screen scraping. It also stores launch scripts, calibration assets, tests and VS Code settings backup/restore tools.
· macro
The working library of instructions that tell AI assistants how to behave, kept in one place instead of scattered across projects.
A private prompt-engineering workspace containing instruction documents, model inventory work, rule templates, planning notes and synchronization concepts. Includes an agent-model inventory scanner; rule-template organisation moved out into the dedicated ai-instruction-sync utility.
The planning and evidence workspace behind everything else: what is being built, what is finished, and the proof it works.
The planning and evidence workspace behind the other projects: what is being built, what is finished, and the proof it works. It contains specifications, task lists, acceptance criteria and the run artifacts that back them, across Kool, the agent systems, the automation lanes and the deployment tooling.
· plans
Keeps one set of AI instructions in step across many projects, so a change written once reaches all of them.
A small CSV-driven utility for keeping agent instruction files consistent across repositories, worktrees and user-level locations. Enabled rows map source templates to destination filenames; dry-run previews changes, real runs copy files and update last-synced timestamps. It includes reusable AGENTS.md/CLAUDE.md templates, diff and audit helpers, legacy maps and archived expansion ideas. It was separated from the broader PromptEngineering workspace.
Rebuilds a complete medical insurance claim from the underlying patient records, and shows its working line by line so every figure can be traced.
An end-to-end proof that a 129-field CMS-1500-style PDT4 claim output can be assembled from normalized patient, insurance, visit, provider, CPT, ICD-10 and claims tables rather than copied from cached strings. It separates deterministic normalization, payer mappings and human review; logs every transformation; supports overrides and review HTML; and uses de-identified data. Current proofs report 517/519 claim rows byte-exact and 531/531 visits byte-identical to the computed source.
Saves conversations with AI assistants as readable, searchable documents, so nothing is lost when a session closes.
A portable chat-archive system that exports conversations from VS Code, Cursor, Claude Code, Codex and Copilot CLI into Markdown. It filters conversations by role, date, size, duplicate/tool-message status and selected sessions; can generate readable titles using local or cloud models; and includes SearchMD utilities for keyword/FTS search and CSV results. IconForge exposes filter_run, SearchMD and generated monthly/summary folders from this system.
Lets you operate an AI coding assistant entirely by voice from a phone: it reads answers aloud and takes dictated instructions.
CodeReader is an Android accessibility application for operating Claude Code without a computer keyboard. It detects completed Claude answers and reads them aloud, accepts an offline Vosk wake phrase and dictated prompts, taps Send, and supports 20 commands for reading speed, navigation and session selection. Controls are available through voice, overlay, notification or Bluetooth headset. The APK is built through a manual javac/d8/aapt2/signing chain without Gradle or Android Studio.
A control room for AI work sessions: see what each one is doing, message it, and track what it is costing.
A cross-engine control layer for Claude Code and OpenAI Codex sessions. The canonical agentcli dispatcher lists sessions, reads transcript tails and sends/resumes messages while handling engine-specific path rules. The repo also contains a chat-manager that keeps unattended sessions alive, monitors models, budgets, escalations and scheduled components, plus audit packets, reviewer/remediation planning, safety ledgers, vendored skills and acceptance tests. IconForge’s llmDash tile launches its chat and usage dashboards.
A private assistant for everyday admin: it reads email, files and messages, drafts replies, and never sends anything without approval.
A private, CLI-first local operations platform for agent handoffs and workflow automation. It connects approved metadata and actions across GitHub, Gmail, WhatsApp exports, Google Sheets/Drive, local files, browser/observation plans, calendars, contacts and phone reports while defaulting to redaction, dry runs, read-only access and human approval. It stores assist events in SQLite and is the live deployment source for extracted Twitter, tenant-mail and Zaps subsystems.
Sends certified physical letters to tenants and tracks them to the doorstep, with rent ledgers and notices generated automatically.
A private tenant-mail application extracted from PersonalOps. It creates mail-merge letters, stages certified Click2Mail jobs, records job state in SQLite, retrieves real USPS tracking numbers and scan timelines, and offers read-only status reports. It also maintains rent ledgers, calculates delinquency/aging, generates dry-run 5-day or 14-day notice plans and renders DOCX letters. Vendor submission is protected by explicit proof, target and environment gates; sensitive addresses and credentials remain outside Git.
Puts an application on a cloud computer and keeps it awake, so anyone with a link can use it without installing anything.
The Kool cloud-server toolchain. It creates and maintains GitHub Codespaces running the Kool frontend, backend and agent runtimes; exposes password-gated public URLs; and supports guided installation, recipient packaging, link sharing, restart-only access, keep-awake controls and delivery. It enforces a no-secrets-in-repo policy and retains historical proof artifacts in the plans repository.
Packages any application so a non-technical person can run their own private copy in the cloud from a single link and password.
A generalized framework derived from the proven Kool Codespaces workflow. It creates/reuses cloud boxes, builds one-window guided installers, packages recipient-specific downloads, supports full-copy, link-only and self-restart sharing tiers, rotates and verifies passwords before delivery, republishes ports, restores services and offers scheduled keep-awake. A template contract separates machine/repo/start layers so other apps can adopt the system. An acceptance harness performs about 85 integrity, parity and secret-leak checks.
Runs a social media account end to end: writes posts, schedules them, replies, and reports what worked. Nothing publishes without a yes.
The Twitter/X subsystem built inside PersonalOps. It reads public/account data through twscrape and performs approved writes through a cookie-injected real Firefox controlled by Playwright, avoiding an official developer API key. Modules cover posting, replies, threads, schedules, engagement, analytics, growth, mentions, CRM, notifications, quote cards and health probes. Publishing is approval-gated, DMs are read-only, and 56 offline tests are documented.
A one-click cloud setup for an AI agent, protected behind a sign-in so only invited people can reach it.
A minimal GitHub Codespaces template that deploys the external OpenClaw agent and its Control UI without the full Kool stack. It configures Node/pnpm/OpenClaw, secrets, public port 18789, allowed origins and token authentication. The project documents real-browser sign-in proof, authenticated API behavior and a completed Gmail-draft task. It is original deployment/integration work around an upstream runtime, not authorship of OpenClaw itself.
Pulls the colours out of any image, shows where each one appears, and recolours the picture while keeping its shading and brushwork.
A local color-palette and image-recoloring studio. It extracts dominant colors while ignoring transparent pixels, displays palette strips, highlights where each swatch appears in the source image, supports eyedropper selection, reordering and rescanning, and performs relative OKLCH color shifts that preserve shading and brushwork. Users can stack swaps, make colors transparent, skin UI regions and export full-resolution PNGs, palette images, JSON, CSS variables and hex lists.
A dashboard showing what is finished and what is still open across every project, with the decisions that need a human.
A control hub for capability matrices, extracted from the planning monorepo into its own program. Its localhost dashboard checks whether project capability matrices are clean or stale, shows shipped versus open roadmap cells, runs sweeps, regenerates generated views, enables or disables a daily scheduled sweep and exposes human approval gates such as voice review and safe Zap dry runs. Configured matrices cover automation, Twitter, Kool optimization, reporting, autonomy and email assistance.
Screens the whole options market each morning and returns a short, explained list of the trades worth looking at.
The repository behind OptionAir, a quantitative vertical-spread screening engine. A one-click run downloads current equity and options data, combines debit and credit structures, evaluates roughly 43,000 spread combinations, calculates about 480 valuation, liquidity, historical and scenario metrics per spread, and scores the survivors under 16 named strategy playbooks. The documented scale exceeds 20 million calculated values per refresh, compressed into an explainable ranked shortlist.
Shows which programs are using which connections on the computer and warns before two of them collide.
A Windows, standard-library developer utility for the fixed-port local application ecosystem. Its registry maps each port to an owning app, role, launcher and command-line signature; a process on a registered port counts as valid only if the signature matches. PortGuard can check apps, refuse conflict-prone launches, scan batch files, discover unregistered listeners and display a live Tkinter dashboard. It covers Kool, Storybook, llmDash, Matrix Program and agent gateways.
Instant search across every saved conversation, so an answer from months ago can be found in seconds.
Search utilities for exported Markdown chat archives. It includes substring/keyword search and an SQLite FTS5 indexed variant with local, git-ignored indexes.
· searchmd
A shared task board where a person and an AI assistant work the same list together, with messages and progress in one place.
Tasker evolved from an always-on-top shared checklist into a Slack-style browser application backed by common board data. Human and agent participants can add, edit, check, reorder and nest tasks; exchange messages; attach files and links; set dates, labels and dependencies; insert templates; search and filter; use split views; inspect activity; and track completion velocity and active versus paused time. CLI parity and delivery bridges allow agents to operate and verify the same workflows.
· tasker
Watches for things happening — an email, a mention — and prepares the response automatically, leaving the send button to you.
A small extracted automation engine implementing POLL → MATCH → ACTION → NOTIFY. It includes scheduling, deduplication, state and kill-switch logic plus handlers for polling sources and creating actions/notifications. Actions are intentionally draft-only and human-gated. Handlers call the PersonalOps Gmail and Twitter modules, where it runs live.
· zaps
Credential evaluation, translation and legal-support services used by thousands of immigrants and hundreds of law firms.
more…Credential evaluation, translation and legal-support services used by thousands of immigrants and hundreds of law firms.
Public-health education, disease prevention, organizational development and nonprofit programming.
more…Public-health education, disease prevention, organizational development and nonprofit programming.
Clinical training and research in headache, sleep disorders and AI-assisted diagnostics.
more…Clinical training and research in headache, sleep disorders and AI-assisted diagnostics.
Volunteer research group focused on clinical decision-making, patient education and AI-assisted diagnostic concepts.
more…Volunteer research group focused on clinical decision-making, patient education and AI-assisted diagnostic concepts.
A confidential online questionnaire that tells someone whether they should get tested for HIV, and prints advice they can hand to a doctor.
more…A confidential online questionnaire that tells someone whether they should get tested for HIV, and prints advice they can hand to a doctor.
HATS is the HIV/AIDS Test Screening Software: a confidential online survey that assesses whether a user should obtain HIV testing, generates customized recommendations, and provides a printable report for a physician or emergency department. The legacy web application is primarily PHP, runs on Apache, stores data in MySQL, and includes PDF reporting and account/report storage. Its documentation also covers Windows and Linux development setup for Medwiser.org contributors.
· hats
Public education, awareness and improvement of sleep-apnea care.
more…Public education, awareness and improvement of sleep-apnea care.
Developed an online sleep-medicine platform and remote clinical workflow.
more…Developed an online sleep-medicine platform and remote clinical workflow.
Led sleep services, grew study volume and launched a home sleep-testing program.
more…Led sleep services, grew study volume and launched a home sleep-testing program.
Clinical neurology and sleep medicine combined with telemedicine, software development and practice automation.
more…Clinical neurology and sleep medicine combined with telemedicine, software development and practice automation.
A nonprofit organization dedicated to improving and advancing medical-marijuana care.
more…A nonprofit organization dedicated to improving and advancing medical-marijuana care.
Places multi-leg options trades through a broker platform, so complex positions go in as one order instead of four.
more…Places multi-leg options trades through a broker platform, so complex positions go in as one order instead of four.
An experimental options-trading codebase for Interactive Brokers Trader Workstation. Buys butterfly structures, submits option orders, returns offset-debit data, and constructs a four-leg combination from a three-leg butterfly plus an offset-debit leg.
A workbench for building and testing AI assistants: set them up, run them against real tasks, and compare which setup performs best.
more…A workbench for building and testing AI assistants: set them up, run them against real tasks, and compare which setup performs best.
Kool.bot (Kool) is the principal AI platform in this portfolio: a Vite/React interface, FastAPI gateway and optimization engine integrating assistants, personas, runtimes, tools and agent systems. Every feature ships with a user-visible integration, a proof test and regression coverage. Documented workflows include project, agent and toolbox configuration, benchmark uploads, optimization pipelines, reports, parameter exploration, output validation, Storybook component previews, and optional OpenClaw, Nanobot and ZeroClaw runtime integrations.
· nia
A shared glossary that keeps everyone using the same word for the same thing across a large project.
more…A shared glossary that keeps everyone using the same word for the same thing across a large project.
A terminology and code-understanding repository centered on a large Project Lexicon. It defines precise meanings, hierarchies and relationships for Kool concepts such as Agent, Persona, Runtime, Tool, Parameter and Project, and records where common names collide with generic dictionary meanings. The goal is consistent technical writing, code generation and review across a complex multi-repository system.
Points at a large codebase and tells you which files actually matter, so a newcomer knows where to start reading.
more…Points at a large codebase and tells you which files actually matter, so a newcomer knows where to start reading.
A software-distillation pipeline designed to identify the essential core of a codebase. It combines structural dependency analysis and centrality, optional runtime coverage, and Git archaeology such as age and churn, then merges the signals into ranked files and directory reports. Expanded from JavaScript/TypeScript to Python, with a staged hybrid funnel in which deterministic analysis precedes LLM interpretation.
· distill
Turns any picture into a Windows icon and builds a one-screen board of tiles that launch every program on the machine.
more…Turns any picture into a Windows icon and builds a one-screen board of tiles that launch every program on the machine.
IconForge is a standalone Tkinter desktop application that converts images to .ico files, creates Windows shortcuts and places them at chosen desktop positions. It also powers the dark Launcher Board shown in the screenshots. The current system includes editable launcher metadata, sections, search, usage statistics, smart folders such as Most Used/recent, portable path tokens, shortcut rebuilding and a Quick Access folder manager. A single registry file keeps every tile pointing at the right program.
Watches long-running AI coding sessions and keeps them moving: spots when one is stuck, restarts it, and queues the next job.
more…Watches long-running AI coding sessions and keeps them moving: spots when one is stuck, restarts it, and queues the next job.
A broad Windows/VS Code automation workspace for supervising AI coding sessions. Covers screen recognition and OCR, chat-title discovery, cycling among visible chats, model switching, CSV task queues, stuck/error detection, restarts, prompt injection, monitoring dashboards, and a CLI runner that drives Copilot through subprocess/JSONL instead of screen scraping. It also stores launch scripts, calibration assets, tests and VS Code settings backup/restore tools.
· macro
The working library of instructions that tell AI assistants how to behave, kept in one place instead of scattered across projects.
more…The working library of instructions that tell AI assistants how to behave, kept in one place instead of scattered across projects.
A private prompt-engineering workspace containing instruction documents, model inventory work, rule templates, planning notes and synchronization concepts. Includes an agent-model inventory scanner; rule-template organisation moved out into the dedicated ai-instruction-sync utility.
The planning and evidence workspace behind everything else: what is being built, what is finished, and the proof it works.
more…The planning and evidence workspace behind everything else: what is being built, what is finished, and the proof it works.
The planning and evidence workspace behind the other projects: what is being built, what is finished, and the proof it works. It contains specifications, task lists, acceptance criteria and the run artifacts that back them, across Kool, the agent systems, the automation lanes and the deployment tooling.
· plans
Keeps one set of AI instructions in step across many projects, so a change written once reaches all of them.
more…Keeps one set of AI instructions in step across many projects, so a change written once reaches all of them.
A small CSV-driven utility for keeping agent instruction files consistent across repositories, worktrees and user-level locations. Enabled rows map source templates to destination filenames; dry-run previews changes, real runs copy files and update last-synced timestamps. It includes reusable AGENTS.md/CLAUDE.md templates, diff and audit helpers, legacy maps and archived expansion ideas. It was separated from the broader PromptEngineering workspace.
Rebuilds a complete medical insurance claim from the underlying patient records, and shows its working line by line so every figure can be traced.
more…Rebuilds a complete medical insurance claim from the underlying patient records, and shows its working line by line so every figure can be traced.
An end-to-end proof that a 129-field CMS-1500-style PDT4 claim output can be assembled from normalized patient, insurance, visit, provider, CPT, ICD-10 and claims tables rather than copied from cached strings. It separates deterministic normalization, payer mappings and human review; logs every transformation; supports overrides and review HTML; and uses de-identified data. Current proofs report 517/519 claim rows byte-exact and 531/531 visits byte-identical to the computed source.
Lets you operate an AI coding assistant entirely by voice from a phone: it reads answers aloud and takes dictated instructions.
more…Lets you operate an AI coding assistant entirely by voice from a phone: it reads answers aloud and takes dictated instructions.
CodeReader is an Android accessibility application for operating Claude Code without a computer keyboard. It detects completed Claude answers and reads them aloud, accepts an offline Vosk wake phrase and dictated prompts, taps Send, and supports 20 commands for reading speed, navigation and session selection. Controls are available through voice, overlay, notification or Bluetooth headset. The APK is built through a manual javac/d8/aapt2/signing chain without Gradle or Android Studio.
Sends certified physical letters to tenants and tracks them to the doorstep, with rent ledgers and notices generated automatically.
more…Sends certified physical letters to tenants and tracks them to the doorstep, with rent ledgers and notices generated automatically.
A private tenant-mail application extracted from PersonalOps. It creates mail-merge letters, stages certified Click2Mail jobs, records job state in SQLite, retrieves real USPS tracking numbers and scan timelines, and offers read-only status reports. It also maintains rent ledgers, calculates delinquency/aging, generates dry-run 5-day or 14-day notice plans and renders DOCX letters. Vendor submission is protected by explicit proof, target and environment gates; sensitive addresses and credentials remain outside Git.
A dashboard showing what is finished and what is still open across every project, with the decisions that need a human.
more…A dashboard showing what is finished and what is still open across every project, with the decisions that need a human.
A control hub for capability matrices, extracted from the planning monorepo into its own program. Its localhost dashboard checks whether project capability matrices are clean or stale, shows shipped versus open roadmap cells, runs sweeps, regenerates generated views, enables or disables a daily scheduled sweep and exposes human approval gates such as voice review and safe Zap dry runs. Configured matrices cover automation, Twitter, Kool optimization, reporting, autonomy and email assistance.
A private assistant for everyday admin: it reads email, files and messages, drafts replies, and never sends anything without approval.
more…A private assistant for everyday admin: it reads email, files and messages, drafts replies, and never sends anything without approval.
A private, CLI-first local operations platform for agent handoffs and workflow automation. It connects approved metadata and actions across GitHub, Gmail, WhatsApp exports, Google Sheets/Drive, local files, browser/observation plans, calendars, contacts and phone reports while defaulting to redaction, dry runs, read-only access and human approval. It stores assist events in SQLite and is the live deployment source for extracted Twitter, tenant-mail and Zaps subsystems.
Saves conversations with AI assistants as readable, searchable documents, so nothing is lost when a session closes.
more…Saves conversations with AI assistants as readable, searchable documents, so nothing is lost when a session closes.
A portable chat-archive system that exports conversations from VS Code, Cursor, Claude Code, Codex and Copilot CLI into Markdown. It filters conversations by role, date, size, duplicate/tool-message status and selected sessions; can generate readable titles using local or cloud models; and includes SearchMD utilities for keyword/FTS search and CSV results. IconForge exposes filter_run, SearchMD and generated monthly/summary folders from this system.
A control room for AI work sessions: see what each one is doing, message it, and track what it is costing.
more…A control room for AI work sessions: see what each one is doing, message it, and track what it is costing.
A cross-engine control layer for Claude Code and OpenAI Codex sessions. The canonical agentcli dispatcher lists sessions, reads transcript tails and sends/resumes messages while handling engine-specific path rules. The repo also contains a chat-manager that keeps unattended sessions alive, monitors models, budgets, escalations and scheduled components, plus audit packets, reviewer/remediation planning, safety ledgers, vendored skills and acceptance tests. IconForge’s llmDash tile launches its chat and usage dashboards.
Packages any application so a non-technical person can run their own private copy in the cloud from a single link and password.
more…Packages any application so a non-technical person can run their own private copy in the cloud from a single link and password.
A generalized framework derived from the proven Kool Codespaces workflow. It creates/reuses cloud boxes, builds one-window guided installers, packages recipient-specific downloads, supports full-copy, link-only and self-restart sharing tiers, rotates and verifies passwords before delivery, republishes ports, restores services and offers scheduled keep-awake. A template contract separates machine/repo/start layers so other apps can adopt the system. An acceptance harness performs about 85 integrity, parity and secret-leak checks.
Puts an application on a cloud computer and keeps it awake, so anyone with a link can use it without installing anything.
more…Puts an application on a cloud computer and keeps it awake, so anyone with a link can use it without installing anything.
The Kool cloud-server toolchain. It creates and maintains GitHub Codespaces running the Kool frontend, backend and agent runtimes; exposes password-gated public URLs; and supports guided installation, recipient packaging, link sharing, restart-only access, keep-awake controls and delivery. It enforces a no-secrets-in-repo policy and retains historical proof artifacts in the plans repository.
A one-click cloud setup for an AI agent, protected behind a sign-in so only invited people can reach it.
more…A one-click cloud setup for an AI agent, protected behind a sign-in so only invited people can reach it.
A minimal GitHub Codespaces template that deploys the external OpenClaw agent and its Control UI without the full Kool stack. It configures Node/pnpm/OpenClaw, secrets, public port 18789, allowed origins and token authentication. The project documents real-browser sign-in proof, authenticated API behavior and a completed Gmail-draft task. It is original deployment/integration work around an upstream runtime, not authorship of OpenClaw itself.
Screens the whole options market each morning and returns a short, explained list of the trades worth looking at.
more…Screens the whole options market each morning and returns a short, explained list of the trades worth looking at.
The repository behind OptionAir, a quantitative vertical-spread screening engine. A one-click run downloads current equity and options data, combines debit and credit structures, evaluates roughly 43,000 spread combinations, calculates about 480 valuation, liquidity, historical and scenario metrics per spread, and scores the survivors under 16 named strategy playbooks. The documented scale exceeds 20 million calculated values per refresh, compressed into an explainable ranked shortlist.
Pulls the colours out of any image, shows where each one appears, and recolours the picture while keeping its shading and brushwork.
more…Pulls the colours out of any image, shows where each one appears, and recolours the picture while keeping its shading and brushwork.
A local color-palette and image-recoloring studio. It extracts dominant colors while ignoring transparent pixels, displays palette strips, highlights where each swatch appears in the source image, supports eyedropper selection, reordering and rescanning, and performs relative OKLCH color shifts that preserve shading and brushwork. Users can stack swaps, make colors transparent, skin UI regions and export full-resolution PNGs, palette images, JSON, CSS variables and hex lists.
Shows which programs are using which connections on the computer and warns before two of them collide.
more…Shows which programs are using which connections on the computer and warns before two of them collide.
A Windows, standard-library developer utility for the fixed-port local application ecosystem. Its registry maps each port to an owning app, role, launcher and command-line signature; a process on a registered port counts as valid only if the signature matches. PortGuard can check apps, refuse conflict-prone launches, scan batch files, discover unregistered listeners and display a live Tkinter dashboard. It covers Kool, Storybook, llmDash, Matrix Program and agent gateways.
Instant search across every saved conversation, so an answer from months ago can be found in seconds.
more…Instant search across every saved conversation, so an answer from months ago can be found in seconds.
Search utilities for exported Markdown chat archives. It includes substring/keyword search and an SQLite FTS5 indexed variant with local, git-ignored indexes.
· searchmd
A shared task board where a person and an AI assistant work the same list together, with messages and progress in one place.
more…A shared task board where a person and an AI assistant work the same list together, with messages and progress in one place.
Tasker evolved from an always-on-top shared checklist into a Slack-style browser application backed by common board data. Human and agent participants can add, edit, check, reorder and nest tasks; exchange messages; attach files and links; set dates, labels and dependencies; insert templates; search and filter; use split views; inspect activity; and track completion velocity and active versus paused time. CLI parity and delivery bridges allow agents to operate and verify the same workflows.
· tasker
Runs a social media account end to end: writes posts, schedules them, replies, and reports what worked. Nothing publishes without a yes.
more…Runs a social media account end to end: writes posts, schedules them, replies, and reports what worked. Nothing publishes without a yes.
The Twitter/X subsystem built inside PersonalOps. It reads public/account data through twscrape and performs approved writes through a cookie-injected real Firefox controlled by Playwright, avoiding an official developer API key. Modules cover posting, replies, threads, schedules, engagement, analytics, growth, mentions, CRM, notifications, quote cards and health probes. Publishing is approval-gated, DMs are read-only, and 56 offline tests are documented.
Watches for things happening — an email, a mention — and prepares the response automatically, leaving the send button to you.
more…Watches for things happening — an email, a mention — and prepares the response automatically, leaving the send button to you.
A small extracted automation engine implementing POLL → MATCH → ACTION → NOTIFY. It includes scheduling, deduplication, state and kill-switch logic plus handlers for polling sources and creating actions/notifications. Actions are intentionally draft-only and human-gated. Handlers call the PersonalOps Gmail and Twitter modules, where it runs live.
· zaps