The Dominio Router
Context engineering for a whole life
I Replaced The Windows Taskbar
Desktop shell built for agents
A Board For AI Sessions
Ops view for many agents
Thread notes system
Durable memory per work item
Production, locked by default
Production writes wait for a human click
Hilos: Threads Hold Memory
Long-horizon memory without auto-memory
A System That Audits Itself
A system that audits its own use
QA Loop With An Adversary
Verification as a habit
My Voice, Coded Once
Personalization by spec
A Catalog For Local AI
Local LLM with guardrails
The Model Reports Its Steps
Agent self-reporting UI
Work Or Personal, Automatically
Context separation
Usage Bar And Tasks Panel
Cost visibility in UI
Consent Tokens For Agents
Consent primitive for agents
A Guard On My Phone
Device-level guardrail
Agent-routing guard
Policy as a hook, not as a hope
Outbound quality stack
No embarrassing auto-messages
A Second Brain In Meetings
A cited card while people talk
A Live Interview Coach
Real-time coach, no API key
Made by Asking, repositioned
Each project separate and sellable
A Home Dashboard For Everyone
Lights, TV and music on one page
Chat history as private memory
Private memory as MCP
My own model, from anywhere
Local Qwen, reachable from any of my devices
One page for the house
Each light controlled from one shared page
Forms filled, values unseen
Forms filled without the AI seeing the data
Fitness data, plumbed together
Three fitness sources joined every day
Credits, only on a go
Spend control on a paid generator
Quiet checks, same job
Less noise, same checks
Each lane, its own browser
Stops cross-lane browser mixups
Sent back, made to try
A give-up reply gets bounced back once
Code work routed to Jules
Uses an included agent automatically
An icon on every tab
Quick visual lane identity
Shell output, compressed
Cuts tokens with no model effort
Agents, fixed-shape reports
Every helper reports verdict first
Done, with proof
Proof before completion
Salesforce work, reviewed
Reusable reviewers, fixed format
Agent browser, logged in
Reliable web automation entry point
Show me, only when asked
Small bridge between agent and person
An AI study page
The next batch follows my votes
Setup, self-explained
The setup explains itself to the model
Software, self-healing
Green commits, red reverts, every 15 s
One shell, every screen
Deep Windows engineering behind the shell
Session shortcuts, keyboard only
Alt plus a number jumps to a session
Phone to PC, same chat
Phone and PC continuity
Heavy apps, governed
Heavy programs slowed down, not killed
Memory, watched
Free monitoring with no model
Plan usage, from my phone
Replaces three remote sessions
My own search engine
Free search backbone
A subscription, put to work
Uses a subscription, not API keys
Sessions, in my pocket
App feel from my own server
Session alerts, on my phone
Phone push when a session needs me
Side panel, from transcripts
Tracks sessions without asking the model
Session summaries, carried over
A readable history of work
Research runner, per tool
Knowing what the subscriptions can do
An API credit watcher
Credit checked every 12 h, warned before expiry
Token budget, always in view
Five-hour and weekly use always on screen
Production, mirrored locally
Production metadata copied nightly, read-only
AI brief, daily at noon
Six researchers, one writer, three judges
AI setup, backed up nightly
Nightly mirror, secret scan before upload
Logins, never seen by the model
Passwords typed by the browser, never seen
Each task, the right account
Every session tagged work or personal at start
Usage meter, on my desktop
Plan use on a bar, green to amber to red
Services, self-restarting
Two services checked every 2 min, restarted
Openings, watched by small jobs
One poller per page, one email on change
Approved messages, one by one
Each message approved before it is queued
Paused work, resumed after reset
Paused work resumes when the plan resets
Internet quality, logged
Ping each minute, speed test every 30 min
AI news, in one email at noon
Capped at 1000 words, sent at noon
The AI week, on Sunday
Up to five ranked fixes every Sunday
The day, in one email
Yesterday's calendar, mail and notes in one email
The channel, every Monday
A verdict per upload batch, every Monday
Broken jobs, caught every hour
Every job's status checked hourly, no AI
Usage warning, once a week
An email a day before reset if under 70%
Reset reminder, sent once
One email, once, only if the reset is unused
My AI use, audited weekly
One CSV row per week from my transcripts
Weekly cleanup, by rule
Old regenerable files deleted weekly by age
Nightly backup, secrets refused
Nothing pushed while a secret is found
Device data, collected daily
Three sources captured at 3:45, each on its own
Today's log, pushed every 30 minutes
Today's numbers published every 30 min, 7:00-23:00
My internet, checked every minute
A ping every minute; the gateway splits home from operator
Internet speed, every half hour
A speed test every 30 min; under 80% of the plan is a dip
The internet week, on Monday
One Monday email; the subject holds drops and dips
Browser change, measured a week later
Click errors compared after seven days
Two browsers, kept in sync
New Edge bookmarks copied to Chrome every 30 min
Festival schedule, watched every 3 hours
Three sources every 3 h; one email, then it expires
Stale services, restarted every 2 minutes
Running and current, checked every 2 min
Several sessions, one resume job
One usage reader every 10 min for every watched session
Live-source rule, graded from logs
Two logs of live-source checks, one report
Pipeline Maker
Agent work as a template
Idea Forge
Idea-to-test factory
Marketing As A Loop
Promo as a measurable loop
A Mode For Not Concluding
Stops forced answers
A Diary With Hard Numbers
Motivation with data
Clear And Start
Context hygiene, one gesture
A Fork Button For Sessions
Branching conversations in UI
Decisions With Previews
Human-in-the-loop UX
Claude Tests Its Own UI
AI testing AI's own surface
Tabs For Notes And Agents
Readable agent output
A Study Feed I Curate
Personal learning loop
Browser fallback hook
Flaky tool, automatic fallback
Guard test suite
Tested guardrails, not trusted ones
Change-control hook
The change request fills itself from the change
Copilot isolation
A leak found in testing, then fixed
Token-waste audit
Evidence-based cost cuts
Freno: token brake
A brake independent of the agent
Plan-limit and reset alarms
No wasted allowances
Model and long-run policy
Cost matched to task length
Fail-cheap rule
Test every idea at its cheapest version first.
Ledger with a hard cap
Budget enforced in code
Subscription as image route
Use what's already paid before paying per call
Auto-intake every 30 min
Weak outputs get redone by a stronger model
Per-item agent launcher
Each agent stops at a draft for approval
Keyword push alert
It pushes the exact line to my phone
Pending-items sweep
Triage with approval
Salesforce MCP, hosted
A dedicated agent user with scoped access
Deal Desk module
Apex logic, 10 flows, an LWC and a dossier PDF
Mass-load builders
Repeatable data loads
Margin rule and cost alerts
A scheduled job flags cost changes
Sharing model rework
Changes shipped as reviewable metadata
Why Users Can't See Fields
Finds invisible fields
Demo Videos From A Script
Demo from a script
User Guides From Code
Docs from code
Every Handoff Looks The Same
Standard handoffs
Month-End In One Run
Month-end in one run
Timesheets From Evidence
Timesheet from evidence
Repo, UAT, PROD: One Table
Repo, UAT and PROD checked in one table
Every Call Comes Prepped
Auto-brief per call
Testing Meeting Tools Offline
Tests prove it never types or clicks
Interview Prep, Reused
Prep reused every call
A Karaoke Teleprompter
Word-by-word delivery
LinkedIn inbox, pre-drafted
A drafted reply for every new message
Applications from a spreadsheet
ATS forms filled from one profile file
Targeting from market data
Demand counted from public job postings
A faceless video factory
Twelve steps, each with its own check
A quality gate for Shorts
Automated quality gate for video
Shorts beat long video
A pivot made from the numbers
Many products, one pattern
Ten product pages from one brief
Finding taste by swiping
Taste capture without vocabulary
An AI Critic For Frames
The AI critic matched me on 7 of 10 clips
Long Renders, Supervised
Supervised long renders
Local vs Hosted Video Models
Local vs hosted decision
The 0.7-Second Rule
Every clip cut to its best 0.7 s
Approving Images, One By One
Approve, reject or comment in one keypress
Claude Drives Blender
bpy automation for any project
A Runbook For Stubborn Bulbs
Bulb pairing steps that worked, written once
A room from one video
Room modeled from video
Claude drives PC and phone
Claude drives both devices
Bookmarks, sorted by script
A separate headless Chrome edits bookmarks
Expenses logged with no typing
Zero-typing ledger
Phone recordings to text
Phone-to-text
Watchers for slow portals
One watcher per portal, each with an end date
Travel planning with agents
Crawl many listings, then filter hard
Buy or rent a GPU?
Rent for bursts, buy for nonstop work
An old phone named Nova
Old phone turned into a home assistant
The TV, one click away
Claude drives the living-room TV
Lights that follow the beat
Lights pulse on the beat of the track
A deck that learns
A swipe deck that learns from each pick
Sandboxes that reset on close
Safe testing of APKs and scripts
Sourced teacher mode
Study with citations
An inspiration board for me
Personal motivation feed
Shopping search through an emulator
App scraping without an API
Google Flow mascot videos
AI video on a subscription
A credit-guarded AI media studio
Credit-capped AI media studio
Idea tree, tested
Ideas tested, branched or killed
One style, every page
Every page shares one stylesheet
Verification codes, on request
Unblocking locked-out users without opening Setup
Agent access that repairs itself
Keeping the agent's sessions alive across systems
Active users, by login
License and usage visibility
IT tickets, house format
Consistent IT Operations tickets without hand formatting
One folder per requirement
A folder per requirement, request to closeout
One topic, fully reviewed
Every claim in the status table has evidence
Rollout advice, per role
Rollout advice per role before production
AI features already paid for
What's licensed is checked before buying
Deploys, guarded
A class without a test stops the deploy
Dog reader, research first
A research dossier before any prototype code
Animation tools, researched
Skills, MCP servers and GSAP mapped, links checked
No proof, no done
No proof after a change, the turn can't end
Salesforce automation, cataloged
What fires on a record, found before building
User stories, on one dashboard
Every story's status on one local page
AI feature, tested for real
Done only when Apex and browser tests pass
Requirements, one server for agents
A phase per requirement, evidence to move on
Agent browser, never on screen
Logged-in Chrome with no window on screen
Laptop, tuned up
Sleep, hibernation and forced restarts off
Work requests, one intake
Chat and email requests in one sheet
Toolbox Routing And Relevo
Multi-vendor orchestration
The Shell On My Phone
Mobile control of agents
Free-tier video models
Free tiers used fully
Contract reader LWC
AI document review in the CRM
Rehearsing on the real platform
Practice where the real interview happens
Walk through a scanned room
3D scanning as a service
An edit, four versions
Iterated editing with automated QA
A Streetwear Brand Mock
A streetwear brand built only as a mock
A Folklore Animation Series
Research to series plan
Three Seconds, One Character
Stopped at round 10 of 12 as over-engineering
Workouts as a pixel game
Wearable data as a game
Coach on the home screen
Native widget to a coach
AI life in a box
Organisms copy themselves and kill for space
Two local AIs, talking
A private thought before each spoken line
Phone control over USB
A route around a blocked input layer
Multi-LLM cost study
Result: use paid tools better first
AI Agents, Live On Stream
Agents debate live, chat steers the topic
Last Lantern: A Postmortem
22 of 458 QA verdicts passed, and why
Stopping LoRA Training Early
Dropped after a stray lantern in 6 of 10 poses