Train agents on the jobs you should not keep doing.
Reports, audits, follow-ups, dashboards, summaries, checks, and recurring analysis become reusable workflows.
Guided self-serve repo or advisory implementation
Building your digital helpers.
Agent OS helps with the work around the "real" work: research, analysis, framing, documentation, presentation, decision support, and the repeatable tasks that drain time before you get to judgment. Paste START_HERE into Cowork, Claude Code, or Codex and the first conversation runs a guided diagnostic that chooses your lane, then builds your vault, dashboard, first agents, and routines through a nine-quest sequence, with watchers and retrieval as optional power tools at the end. The whole setup is built to feel like a forward-deployed engineer working beside you.
Free and public. The repo is the whole kit; click above to get it.
You get range, not just automation.
Repetition is the easiest doorway. The larger promise is helping you take on complex work that used to cost too much time, attention, or mental load.
Send agents ahead when the scope is unclear.
They gather sources, compare options, create structure, and give you a first artifact to refine.
Let the system write the trail as it works.
Research, decisions, corrections, costs, and handoffs become memory instead of extra admin work.
Keep people in the highest-value seat.
You decide what matters, what will land with humans, and which recommendation is actually worth pursuing.
Open the parts that match your work.
Leverage: what your personal digital team does
Gather and frame
Agents collect inputs, read context, rank options, and turn loose ideas into something reviewable.
Analyze and package
They compare paths, find risks, draft summaries, and create pages, decks, dashboards, or briefs.
Remember and improve
The vault, retrieval, and save-ups keep lessons, decisions, corrections, and next steps available.
Why it matters: the human does the real work
Less personal compute
You do not burn the best part of your attention on setup, formatting, or rebuilding context.
More surface area
You can explore more ideas, learn faster, and reach a useful yes or no sooner.
Better judgment
You stay focused on taste, strategy, relationships, and the real-world context agents do not have.
Setup path: how the promise turns into a system
Start with the diagnostic
The co-pilot asks what kind of work you do, then recommends Cowork, Claude Code, Codex, or a blended setup.
Build memory first
Vault, handoffs, and lint keep the system from becoming another forgetful chat thread. Retrieval comes later, once the vault has enough worth searching.
Add routines and dashboard
Recurring jobs, status, costs, blockers, optional watchers, and agent ownership become visible instead of scattered.
Start with familiar jobs, then take on the bigger lift.
The first wins are practical. From there, Agent OS can help with research, strategy, analysis, and presentation work that used to be hard to start.
Weekly performance report
Agent job: gather metrics, spot movement, write takeaways, and format a client-ready deck.
Monthly budget review
Agent job: consolidate exports, flag changes, explain drivers, and produce a summary.
Pipeline follow-up
Agent job: review open opportunities, draft next steps, and surface stalled deals.
Meeting and task cleanup
Agent job: summarize notes, extract owners, update open loops, and prepare the next brief.
Account check-in
Agent job: read history, identify risks, draft a status update, and recommend next actions.
Grant-fit review
Agent job: gather official sources, compare programs, rank fit, write recommendations, and package a review page.
Productization roadmap
Agent job: turn a rough idea into a structured map, surface missing pieces, and create a visual artifact for review.
Story and launch planning
Agent job: organize references, build scene packets, draft options, track feedback, and preserve the creative trail.
The human stays in the highest-value seat.
The point is not to remove people from the work. It is to move them toward judgment, taste, relationships, and real-world decisions.
Let agents handle the material.
Research inputs, organize notes, run audits, update dashboards, compare programs, draft options, format pages, write first-pass presentations, and document what changed.
Keep humans focused on taste and consequence.
You decide which idea matters, what the real-world context changes, how it should land with colleagues or clients, and which option is actually worth pursuing.
The journey is a ladder.
Each step adds leverage without forcing you to understand the whole system on day one.
1
Ask in chat
Use AI for a single question, draft, summary, or idea.
2
Save the context
Store useful files, examples, decisions, style rules, and working notes in the vault.
3
Train the pattern
Turn repeatable jobs and preferred output shapes into routines the agent can run again.
4
Explore bigger work
Use agents for research, comparison, planning, analysis, and first-pass artifacts.
5
Coordinate agents
Give different agents different jobs and prevent duplicate work.
6
Operate the system
Track status, costs, context, handoffs, and risks in a visible operating layer.
What you get that prompting alone can't.
The unique value is turning AI use into an operating system with memory, continuity, quality control, spend awareness, and a way to keep complex work moving.
The vault keeps history
Prior work, decisions, files, examples, and handoffs become reusable context.
The agent finds the right context
An optional local search index helps the agent pull relevant memory instead of asking the user to re-explain. No API keys needed.
Save-ups prevent cold starts
Long-running work can continue across sessions without losing the plot.
Lint and evals protect quality
Vault Lint, checks, examples, and evals keep the system connected and trustworthy.
A local dashboard shows the work
A local HTML dashboard tracks setup progress, agents, and open loops from the first conversation. Connect Notion only if you want it.
Agents expand your surface area
You can investigate more ideas, learn faster, and get to a good or bad decision with less manual setup.
The system writes the trail
Research, decisions, corrections, and next steps become memory instead of disappearing after the session.
People guide the final call
The agent does the prep and structure. The human brings taste, context, relationships, and consequence.
Five profiles tune the path
Organizer, Deep Worker, Enterprise Operator, Builder, or Team Lead: the diagnostic adapts pacing, examples, and lane to how you actually work.
Skills teach preferred outputs
The agent learns how you want summaries, reports, decks, QA, and follow-up to look.
One kit, any domain
The same templates, boot prompts, routines, and quality gates adapt to law, real estate, creative, or operations work. Verticalized packs are on the roadmap; the foundation ships today.
The diagnostic is the first conversation.
Paste START_HERE.md into Cowork, Claude Code, or Codex as your first message. Nine questions decide your lane, then the co-pilot writes the plan and builds through a Q0 to Q8 quest sequence, with a progress file and a local dashboard tracking every step.
What should it help with first?
The build anchors in a real use case, not a demo, and ends with a test that proves it worked today.
Which profile fits you?
Organizer, Deep Worker, Enterprise Operator, Builder, or Team Lead. The profile sets pacing and examples.
How often do you use the terminal?
Never, rarely, sometimes, or every day. The answer scores your lane (Cowork, Claude Code, Codex, or a blend), and the score outranks stated preference.
What exists already?
Where you already use AI, your operating system, and any Obsidian, Notion, or project folder you care about, so nothing gets overwritten.
What are the boundaries?
Data and repos the system must avoid. Then the co-pilot proposes your first two agents and asks you to confirm or correct.
One stack, multiple entry points.
The diagnostic recommends the entry point that fits how you want to work. Cowork, Claude Code, Codex, and blended setups can all build the same memory-backed system.
Cowork guided teammate setup
Cowork can run the full setup flow. It feels most like a conversational digital teammate that explains choices, keeps you oriented, and guides the early build.
Primary jobBuild the same Agent OS foundation through a chat-first setup flow.
Good forPeople who want a digital teammate feel and clear guidance.
Watch forSome local execution still needs user approval or a watcher.
OutcomeVault, dashboard, first agents, optional watchers, evals, and save-up rhythm.
The diagnostic turns into a build map.
The co-pilot uses markdown playbooks and a few safe scripts to implement each layer in your own workspace.
Memory-backed agent operating system
Agents preserve context, retrieve prior work, coordinate through shared files, track progress, and run safety checks.
The memory system can expand only when needed.
Start with the plain-English version. When you want the deeper mechanics, open the stack and see how agents retrieve context, preserve continuity, and stay useful across sessions.
How a question becomes an answer
1. You ask
A prompt lands in Cowork, Claude Code, or Codex.
2. Agent scopes
The agent decides which facts are needed: current project, rules, recent messages, or historical context.
3. Retrieval searches
With the optional index built, the agent searches it before broad file reads and gets likely source files with snippets.
4. Files are read directly
After retrieval points to a source, the agent reads the specific file or section.
5. Workspace state is checked
Fresh files, tests, git status, and your latest instruction can override older memory.
6. Answer lands
The response combines your prompt, retrieved memory, direct file reads, and live workspace state.
What lives in the memory stack
Obsidian vault
The human-readable source of truth for project notes, architecture, decisions, contacts, and session history.
BRAIN_INDEX
The boot map that tells agents what exists, where key files live, and what state must be preserved.
Vector retrieval
An optional local index so agents can find relevant memory by keyword match. Semantic search is the documented upgrade path.
VaultBus
Status files, command notes, escalations, and events move work between agents.
Onboarding progress
A progress file plus a local HTML dashboard keep setup state visible in plain English.
Session handoffs
Save-ups turn long conversations into next-version operating packets.
What keeps the memory trustworthy
Vault Lint
Checks structure so important notes do not become disconnected or invisible.
Evidence gates
Security, release, and quality claims need scan output, test results, diffs, or verifier logs.
Privacy boundaries
Credentials stay in environment variables or Keychain. Private chat and inbox content stay out of public collateral.
Known gaps
Retrieval is optional and keyword-based out of the box. Semantic search and richer aggregation are documented upgrade paths.
Evals
Mistakes become instruction changes, then get retested against future work.
Routines
Recurring lint, retrieval refresh, security sweep, and save-up checks keep the system healthy.
The repo includes a practical Skills Pack.
This is what you get by implementing the starter kit: operating routines that make agents remember, check, hand off, and improve.
Vault setup
Shared folders, BRAIN_INDEX, sessions, decisions, handoffs, and VaultBus coordination.
Local dashboard
A default HTML quest tracker, with Notion as an optional command-center-style layer.
Save-ups
Agents write handoffs and next-session prompts before context quality drops.
Retrieval
A local starter routine so agents can find approved vault context without bulk-reading.
Vault Lint
Checks links, orphans, generated files, and structural issues so the vault stays usable.
Security sweeps
Token scans, history checks, watcher boundaries, and public-release gates.
Evals
Capture friction, lessons, patches, and retests so the setup improves after breakdowns.
Blended ownership
Clear lanes for Cowork, Claude Code, Codex, and hybrid agent teams.
The public repo is the implementation library.
Vault setup
Create the folder structure, BRAIN_INDEX, session registry, and handoff lanes.
Vector retrieval
Optional local index and search scripts. No dependencies and no API keys out of the box.
Vault lint
Run quality checks so the vault stays usable as it grows.
Agent evals
Review agent output, friction, setup quality, and next improvements.
Security agents
Set up sweeps for token families, API keys, risky scripts, and release gates.
Token sweeps
Scan working tree and git history before anything becomes public.
Progress dashboard
Create the local quest dashboard first, then add Notion if you want a coordination hub.
Routines
Define recurring maintenance, security, lint, and status checks.
Two ways to use it.
Use the guided repo yourself, or reach out to Yacob for advisory services to adapt the same setup to your team, tools, and workflows.
Guided self-serve repo
Use the public Agent OS Starter Kit with a setup co-pilot that walks through the decisions, files, and scripts like a forward-deployed engineer by your side.
- Free public instructions and templates.
- Diagnostic path for Cowork, Claude Code, Codex, or a blend.
- Your local vault and your tools.
- Brain setup, retrieval, save-ups, and optional watchers.
- Default local HTML dashboard.
- Optional Notion coordination hub.
- No private Command Center internals.
Advisory implementation
Work with Yacob to design the polished version for your work, team, or company when you want an expert implementation partner.
- Bespoke stack architecture.
- Command Center design and operating model.
- Agent roles, routines, evals, and security review.
- Training on how to manage the system.