Tools built around real work, with the decisions left visible.
AI system · Mac app available as a prerelease.Choose an AI model for a task, then check its answer.
Topics: Agents & permissions · Evidence & evaluation
Available: Source code · Screenshots
- Problem
- Choosing a model and deciding how much effort a task needs can become a task of its own.
- Built
- A Mac app that routes tasks across a chosen set of AI models, reviews their answers, and makes a limited number of corrections when needed.
- Design decision
- I kept the routing choice and review checks visible, with a limit on automated correction attempts.
- Where it stands
- This sample produced the requested guide; two automated checks remained uncertain. A generated answer still needs human judgment.
- Take a look
- Browse the source code, download the Mac app, and explore its routing checks on GitHub.
Explore the app and source code ↗
View visual references ↗
Product build · Desktop and server versions available in alpha.Keep the moving parts of a school music program in one place.
Topics: Workflows & operations
Available: Source code · Screenshots
- Problem
- Directors juggle students, instruments, checkouts, events, and records across disconnected systems.
- Built
- A desktop app and self-hosted option that bring a music program’s day-to-day work together.
- Design decision
- I brought students, instruments, checkouts, and events into one working record system, with desktop and self-hosted options.
- Where it stands
- The alpha releases and sample screens are available to inspect. Reliability in a school’s day-to-day use still needs evaluation.
- Take a look
- Explore the source code, sample-data screenshots, and desktop and server releases.
Explore Band Office ↗
View the visual reference ↗
Personal operating system · Local operating system · related sample files availableA file-backed system for planning work, keeping decisions, and making projects easier to resume.
Topics: Agents & permissions · Workflows & operations
Available: Sample files
- Problem
- Long-running work needs a place for priorities, approvals, and current project context.
- Built
- A local operating layer with structured project records, work queues, handoffs, and a supporting cockpit.
- Where it stands
- These public sample files illustrate the operating-file structure. They are not the private Desk cockpit or its live work queues.
- Take a look
- Inspect the fictional operating files included with the related AgentOS Doctor tool.
Inspect the related operating-file example ↗
Review workspace · Local review workspace · screenshot and sample input availableReview one evidence-backed record at a time, with a clear history of human decisions.
Topics: Evidence & evaluation · Workflows & operations
Available: Screenshots · Sample files
- Problem
- When source material changes, a prior decision may need to be reconsidered.
- Built
- A local review workspace that preserves intake snapshots, records reviews, and produces exports after human approval.
- Design decision
- I preserved source snapshots and human review decisions separately so a changed record can be reconsidered.
- Where it stands
- This example shows intake and review controls, rather than a completed approval or export. The practice build is independent of the live job-search system; all names, URLs, and salaries in the sample are fictional.
- Take a look
- Inspect the review interface and fifty fictional job records from the practice build’s deterministic generator, including missing fields and deliberately invalid examples.
Inspect the fictional intake sample ↗
View the visual reference ↗
Routing utility · Public source packageA small utility for choosing a model route and keeping a record of the choice.
Topics: Agents & permissions
Available: Source code · Sample files
- Problem
- Routing decisions need consistent rules and a way to explain why a route was chosen.
- Built
- A provider-native routing policy with capability and effort recommendations, conservative risk flags, and metadata-only receipts that distinguish a planned route from witnessed execution.
- Where it stands
- The package is advisory and experimental. It does not switch an active root conversation, proxy traffic across providers, or establish model quality or savings.
- Take a look
- Browse the public source package, documentation, and available examples.
Explore the public repository ↗
Inspect the public examples ↗
Data-model toolkit · Local toolkit · sample JSON availableDefine and check the structure of an application’s data before building its interface.
Topics: Workflows & operations
Available: Sample files
- Problem
- Unclear tables, linked records, and field rules can make an application difficult to change.
- Built
- A toolkit for validating tables, records, linked fields, and views, with sample schemas for practical workflows.
- Design decision
- I made field and record rules inspectable before building screens on top of them.
- Where it stands
- These are data-contract examples. The record’s example.com link and sample score are placeholders.
- Take a look
- Open the original Proof Library schema and one fictional record. The schema defines required fields, choices, a filtered view, and sorting.
Inspect the sample schema ↗
Inspect the sample records ↗
Application planning toolkit · Local planning toolkit · sample JSON availableTurn a checked data model into an inspectable application plan.
Topics: Workflows & operations
Available: Sample files
- Problem
- A data schema does not explain which screens people need or how they will use them.
- Built
- Structured application templates and screen descriptions that extend the TableForge data contract.
- Design decision
- I carried field bindings, view filters, and sorting from the data contract into an explicit screen plan.
- Where it stands
- The output is an application plan in JSON. It does not generate a completed frontend or deployed application.
- Take a look
- Inspect the original template and screen manifest produced for the same Proof Library data contract.
Inspect the application template ↗
Inspect the screen manifest ↗
Business workflow system · Local build · sample schema availableOrganize business contacts, companies, and sales work in a shared application.
Topics: Workflows & operations
Available: Sample files
- Problem
- Client and prospect information becomes difficult to use when spread across separate notes and tools.
- Built
- A client-management system with structured contacts and company records, plus focused workflow tools.
- Where it stands
- This reference is a data contract for the CRM workflow, not a screenshot of customer records or proof of a production deployment.
- Take a look
- Inspect the original TableForge data contract for companies, contacts, opportunities, activities, and tasks.
Inspect the CRM data contract ↗
Substitute lesson-planning tool · Live web toolPrepare an emergency substitute lesson plan for a chosen subject, grade, and class period.
Topics: Learning & enablement
Available: Live site · Screenshots
- Problem
- A substitute needs usable teaching instructions when a teacher has little time to prepare.
- Built
- A web tool with subject, grade, duration, substitute-experience, topic, and optional Tennessee-standards inputs, plus options for a read-aloud script and student activity sheet.
- Where it stands
- The live interface is available. This review did not evaluate generated lessons or establish school-wide use or learning outcomes.
- Take a look
- Open the live planning form and inspect the interface screenshot.
Open Sub-Save ↗
View the visual reference ↗
Local publishing utility · Local generator · generated example availableTurn a small source-data file into a printable morning newspaper.
Topics: Workflows & operations
Available: Sample files
- Problem
- A useful personal brief needs a readable structure before adding integrations or automation.
- Built
- A local generator that normalizes manually supplied JSON into newspaper sections and printable HTML.
- Where it stands
- This example uses manual fictional JSON. Live email, calendar, and automatic printing integrations are not demonstrated.
- Take a look
- Open a newspaper generated by the original CLI from the supplied fictional workshop input.
Read the generated sample newspaper ↗
Inspect its fictional source input ↗
Interactive portfolio · Published portfolio · 3D and readable viewsExplore AI implementation and enablement work through a connected 3D museum.
Topics: Websites & portfolios
Available: Live site · Screenshots
- Problem
- A portfolio needs to show both the work and the person behind it, while giving visitors a quick way to find relevant examples.
- Built
- A browser-based museum with six themed exhibits, an illustrated visitor, a retro desktop catalog, a playable resume, and readable project pages.
- Design decision
- I made exploration optional: a project plaque opens the actual work, and visitors can use the Archives, readable pages, or resume directly.
- Where it stands
- The museum uses original exhibit concepts, generated artwork and a textured visitor model, rendered in Three.js. The 3D experience is heavier than the text view.
- Take a look
- Explore the museum, browse its project catalog, or use the text view.
Explore the museum ↗
Browse the readable catalog ↗
View the visual reference ↗
Operating-file diagnostic · Public source packageCheck whether an AI operating system’s files, state, dashboards, and governance have drifted apart.
Topics: Agents & permissions · Evidence & evaluation
Available: Source code · Sample files
- Problem
- A long-running operating system can keep showing inconsistent priorities or stale state across its files.
- Built
- A report-only skill for Codex and Claude Code, with configurable operating-file checks and a fictional sample workspace.
- Where it stands
- The diagnostic reports mismatches for human review. It does not edit files, choose priorities, or change authority.
- Take a look
- Browse the public source, documentation, and included examples.
Explore the repository ↗
Inspect the fictional operating-file examples ↗
Software handoff toolkit · Public source packageKeep a small tool’s purpose, launch command, access boundaries, and health check beside it.
Topics: Releases & handoffs
Available: Source code · Sample files
- Problem
- Months after a small tool was built, its purpose and safe operating conditions can be difficult to recover.
- Built
- A plain capsule.json manifest and Python CLI for manifest validation, path checks, and readable health reports.
- Where it stands
- Health commands run only when explicitly requested. A passing report is not a security audit or a guarantee that the software is safe.
- Take a look
- Browse the public source, documentation, and included examples.
Explore the repository ↗
Inspect the example capsule ↗
Permission-aware AI workbench · Public source packageAnswer a customer question using permitted company sources, then require approval before acting.
Topics: Agents & permissions · Workflows & operations
Available: Source code · Screenshots · Visual preview
- Problem
- A useful answer needs to respect role access, and a draft must remain separate from permission to send it.
- Built
- A React workbench and Python backend with role-filtered sources, citations, owner approvals, local dry-run outbox artifacts, and event receipts.
- Where it stands
- The company and customer are fictional. Role selection is a demonstration control, not authentication; production identity, multitenancy, and live email sending are out of scope.
- Take a look
- Browse the public source, documentation, and included examples.
Explore the repository ↗
Watch the recorded workflow ↗
View the visual reference ↗
Content workflow system · Local workflow build · screenshot availableMove content from an initial idea through drafting, review, and planning.
Topics: Workflows & operations
Available: Screenshots
- Problem
- A draft and an approved version need to remain distinguishable as content changes.
- Built
- A content workspace with revision-specific approval and a persistent record of work.
- Design decision
- I tied approval to a particular revision so an edited draft would not silently inherit a previous approval.
- Where it stands
- This reference shows the local demo interface. It does not demonstrate an automated publishing run or publishing results.
- Take a look
- Inspect the demo dashboard and its separate queues for shaping, approval, review, and planning.
View the visual reference ↗
Workflow demonstration · Local demonstration · screenshot availableExplore AI-supported implementation work through a fictional field-operations company.
Topics: Learning & enablement · Workflows & operations
Available: Screenshots
- Problem
- Implementation teams need answers, readiness decisions, and review points tied to supplied source material.
- Built
- A synthetic-data demonstration with source-based answers, readiness briefs, recommendation review, and an audit history.
- Design decision
- I separated source-backed answers, readiness recommendations, human approval, and the audit history into distinct stages.
- Where it stands
- This build uses synthetic documents, deterministic retrieval and generation, and local browser state. The screenshot demonstrates the workflow, not customer deployment.
- Take a look
- Inspect the dashboard screenshot from the fictional Northstar FieldOps demonstration.
View the visual reference ↗
Hiring workflow demonstration · Local build · screenshot availablePrepare hiring materials while keeping candidate decisions with a human reviewer.
Topics: Workflows & operations
Available: Screenshots
- Problem
- Hiring support needs to distinguish supplied candidate evidence from generated drafts.
- Built
- A fictional-role demonstration with a hiring brief, duty scorecard, candidate packet, interview guide, debrief, and communication drafts. Its agent behavior is simulated.
- Where it stands
- Agent behavior is simulated. The sample is a hiring workflow demonstration, not an automated candidate-selection system.
- Take a look
- Inspect the original interface capture.
View the visual reference ↗
Planning-tool suite · Local build · screenshot availableExplore workflow opportunities, adoption planning, and workforce readiness together.
Topics: Learning & enablement · Workflows & operations
Available: Screenshots
- Problem
- A promising AI task also needs stakeholder planning, review, and preparation for the people using it.
- Built
- Three planning demonstrations: Automation ROI Console, AI Adoption Playbook, and AI Training Diagnostic. Scoring and planning are deterministic; projected savings are not demonstrated results.
- Where it stands
- These three frontend tools use deterministic planning and scoring. They do not establish adoption, customer use, or measured savings.
- Take a look
- Inspect the original interface captures.
View visual references ↗