Personal agents and OpenClaw
Testing how small specialized agents can coordinate, remember, and help with real work.
Personal agents / private apps / field notes
MJGIVAI
Malcolm Graham's working lab for personal agents, private app builds, OpenClaw experiments, secure AI systems, and practical web tools that prove what is actually useful.
Now
Current signal
Things on the bench.
Building
Exploring
Collecting
Daily Agenda project history
Recording how a private agenda app moves from local signals to TestFlight, including the mistakes and privacy lessons along the way.
Read the build historyWeb experiments worth studying
Interactive explanations, useful personal sites, and interfaces that make the web feel handmade again.
Field Notes
Notes
Technical notes with room for side quests.
A practical personal Agent OS
A working model for memory, permissions, handoffs, specialist agents, and the line between automation and trust.
Building a private agenda app from scattered signals
A public, sanitized record of the app's first TestFlight loop, Health and Messages obstacles, review gates, and privacy fixes.
Secure personal AI stack
A plain-language checklist for local-first tools, credential hygiene, agent permissions, and public deployment habits.
Agent OS
OpenClaw / agents
A practical personal Agent OS.
A useful personal agent should remember enough to be continuous, ask before risk, hand work to the right specialist, and automate only inside clear trust boundaries.
Pancho keeps the thread.
Pancho acts as coordinator, dispatcher, and summary point. Doozer builds, Einstein researches, Patton handles operations, Leonardo reviews design, and Rosco audits security.
Daily memory
Scoped handoffs
Ask before risk
Read-first automation
What exists, what is missing, and why.
Exists: daily memory, role files, specialist agents, cost limits, credential rules, and conservative automation.
Now: long-term `MEMORY.md` distills daily notes into durable shared rules.
Next: keep curation lightweight so durable preferences stay aligned with how the agents actually work.
Who handles what.
The stack works best when each agent has a clear job and Pancho keeps decisions, context, and final handoff in one place.
Pancho
Owns context, routes work, checks risk, and brings decisions back to Malcolm.
Doozer
Implements bounded changes, keeps patches scoped, and reports changed files.
Einstein
Finds sources, compares options, and turns fuzzy questions into usable briefs.
Patton
Handles systems, deployment posture, runbooks, and operational cleanup.
Leonardo
Reviews mobile-first layout, visual hierarchy, tap targets, and presentation quality.
Rosco
Audits code, dependencies, credentials, local exposure, and public-service risk.
Secure Stack
Field guide / security
Secure personal AI stack.
A practical checklist for keeping personal agents useful without letting convenience outrun memory, permissions, credentials, or deployment discipline.
What the guidance says in plain English.
OWASP's LLM guidance warns about prompt injection, data leakage, excessive agency, and unsafe tool access. NIST frames AI risk as governance, measurement, management, and transparency. CISA keeps the operational bar simple: design secure defaults, reduce exposed attack surface, and make risky actions intentional.
-
01
Local context
Files, memory, and notes stay in the private workspace first.
-
02
Agent work
Tools run with least authority and narrow task context.
-
03
Permission gate
Risky, public, costly, or account-level actions require review.
-
04
Review loop
Tests, diffs, and audit checks happen before public deployment.
-
05
Deploy public
Only approved and verified changes leave the private workspace.
Keep private context local first.
Put durable preferences and operating rules in workspace files. Avoid copying private memory, credentials, or raw chat context into public pages, prompts, commits, or issue comments.
Use narrow credentials.
Prefer existing authenticated sessions, password managers, short-lived tokens, and per-service scopes. Never ask the user to paste raw passwords into chat.
Gate agent actions by risk.
Reading, organizing, and local edits are low-risk. Spending money, changing DNS, sending messages, touching auth, or publishing externally should require explicit approval.
Review before public deployment.
Check diffs, run the smallest meaningful tests, scan secrets, verify mobile layout, and confirm the live production page after deployment.
Where Malcolm's stack already lines up.
Strong match: local workspace memory, Pancho as coordinator, specialist handoffs, explicit cost limits, and conservative rules for external actions.
Main gap: long-term memory curation should be kept current so old preferences do not drift away from how the system actually works.
Reasonable difference: this is a personal stack, not an enterprise AI program, so the right control is lightweight review and narrow authority rather than heavy policy ceremony.
Experiments
Field notes / tools
Working notes that can turn into useful tools.
A practical library for the things worth testing in public: personal agents, small infrastructure, secure AI workflows, and cheap web systems that are useful without becoming a maintenance project.
01
Personal Agent OS map
A visual map of Pancho, specialist agents, memory, permissions, and safe automation inside this OpenClaw workspace.
Live note02
OpenClaw setup cost calculator
A planning tool for the upfront and monthly costs behind a practical personal agent setup, now expanded into a shareable cost guide.
Live tool03
Secure personal AI stack
A plain-language checklist for local-first tools, credential hygiene, agent permissions, and public deployment habits.
Live guide04
Model Usage Map
A compact read on public AI model usage signals: survey selections, builder demand, and live router traffic.
Live note05
AI Origin Lab
A local-only signal check for text and images. It cannot prove origin, but it shows patterns worth inspecting.
Live tool06
AI readiness scorecard
A quick readiness check for teams deciding whether they are still organizing basics, ready for a contained pilot, or ready for useful automation, with a local pilot brief builder.
Live tool07
Automation Menu
A practical catalog of business workflows that can become small, secure, reviewable automations.
Live guide08
Daily Agenda build history
A public case study about turning local context into an iPhone agenda app while keeping private data out of the published story.
Live storyBuild Log
Lab notebook
What changed, why it matters.
Updated July 5, 2026 after the Fable UX pass.
Short operating notes from the workbench: shipped site changes, security decisions, design reviews, and lessons that are useful beyond one deploy.
Daily Agenda build history goes live.
The site now includes a sanitized case study of the Daily Agenda app's path to TestFlight, including HealthKit fixes, local Messages permissions, multi-agent review, and the decision to remove private bundled agenda data before release.
Website previews are phone-first.
Local site review now starts with a phone-accessible preview URL, not a desktop-only localhost link, because mobile is the first review surface for new MJGIVAI website material.
Automation menu added.
The homepage now translates readiness and pilot planning into concrete workflow options with required inputs, effort, guardrails, and useful first-version outputs.
AI readiness scorecard added.
The homepage now includes a local-only assessment that turns workflow, data, security, and ownership signals into a plain next-step recommendation and a one-page pilot brief.
Comment system hardened and cleaned up.
Production Turnstile uses the real domain widget, admin pages are kept out of crawler paths, and smoke-test comments were removed from the moderation queue.
Agent Stack Map becomes a public explanation.
The site now names the agent roles directly so visitors can understand the operating model behind the tools and notes.
OpenClaw cost calculator goes live.
The calculator turns a personal-agent setup into editable one-time and monthly costs, then invites moderated discussion.
Project Story
Daily Agenda / July 2026
Building a private agenda app from scattered signals.
Daily Agenda started as a practical question: can a personal agent turn local emails, meetings, notes, conversations, and health context into a useful day plan without exposing the private data that makes it useful?
Internal TestFlight, real sources, private by design.
The app is now at TestFlight build 0.2.0 (13). It has a SwiftUI dashboard, Apple Health sync, Messages and Outlook source wiring, a local production pipeline, and a review process that treats useful personal context as sensitive by default.
0.2.0 (13)TestFlight build
PrivateNo bundled agenda data
Best effortiOS background refresh
Seven steps from prototype to phone.
- Started with a local agenda pipeline. The first version pulled together Gmail, Plaud, Apple Notes, Notion, and calendar context into a daily operating brief.
- Moved from report to product. The static output became a SwiftUI dashboard with a command center, calendar rail, source mix, and task cards built for quick scanning on an iPhone.
- Made release repeatable. Xcode signing, App Store Connect uploads, simulator tests, and release gates turned the work into an internal TestFlight loop.
- Added Apple Health carefully. The app now reads steps, exercise minutes, and stand minutes, stores only an aggregate snapshot, and refreshes hourly while active with best-effort background refresh.
- Unlocked Messages with local permissions. Once macOS privacy access was granted, the pipeline could use local Messages context safely and surface only conservative actionable follow-ups.
- Ran specialist reviews before shipping. UI, build, security, operations, and sync behavior each got a focused pass so the release was checked from more than one angle.
- Removed private seed data before TestFlight. A generated agenda bundle was useful for demos, but it contained real personal and work context. We removed it and shipped build 12 without a private agenda snapshot in the app bundle.
The phone build started as the wrong app.
Early TestFlight work exposed a demo shell where the production dashboard should have been. We replaced it with the real ContentView, then made archive and simulator checks part of the release habit.
HealthKit looked broken when the answer was zero.
HealthKit can return no data for a valid day. The fix was to treat no-data responses as zero, persist the last aggregate snapshot, and keep refresh behavior honest.
macOS protected local Messages access.
Full Disk Access and assistive permissions had to be handled before the pipeline could use local Messages context. The ingest now fails softly instead of breaking the whole agenda.
A useful demo seed became a security risk.
The app briefly bundled a generated agenda snapshot with real personal context. The security review caught it, the resource was removed, and the next version shipped without private bundled data.
Model Map
Usage signals / not a census
A map of where AI model attention is going.
There is no clean public pie chart for global model usage. This section compares three useful signals instead: what developers say they use, what AI builders are considering, and where OpenRouter traffic was moving on the captured date.
Stack Overflow top model-family mentions, normalized.
Map,
not census
not census
OpenAI GPT 36.9%
Claude Sonnet 19.4%
Gemini Flash 16.0%
OpenAI Reasoning
15.7%
OpenAI Image 12.1%
Stack Overflow responses are multi-select, so this chart shows share of reported selections among the top categories, not share of developers and not market share.
Families builders are using or considering.
Artificial Analysis surveyed AI practitioners. This is useful for vendor shortlists, but it still measures demand and consideration, not total production volume.
Latest token-routing attention.
- Claude Opus 4.7#1
- Claude Sonnet 4.6#2
- Gemini 3 Flash#3
- DeepSeek V3#4
- Kimi K2#5
- Gemini 2.5 Flash#6
- MiniMax M2#7
- GPT-5#8
OpenRouter is a live routing ecosystem. These ranks move quickly, so this snapshot should be read as dated momentum on that platform only.
How to read it.
OpenAI remains the broad default in surveys. Claude is especially strong around coding and agent workflows. Gemini is prominent in fast, cost-sensitive, and Google-stack work. DeepSeek, Kimi, MiniMax, Qwen, GLM, and Llama matter for open-weight, regional, and cost-sensitive deployments.
Scorecard
Client-side tool
AI readiness scorecard.
A short, local-only check for whether a workflow is still in the basics stage, ready for a contained pilot, or ready for useful automation. Add a few project details and it drafts a one-page pilot brief in the browser. No account, database, or API call is involved.
Turn the score into a first plan.
Keep the first pilot narrow: one workflow, one owner, one success measure, one review gate, and one clear reason to stop.
Cost Lab
Interactive note
OpenClaw setup cost calculator.
Estimate the one-time setup and monthly run-rate for a practical personal agent stack. Defaults are planning assumptions, not vendor quotes, so every line stays editable. MCP services are modeled as optional connector costs, not a dependency of this static page. The expanded guide is available at the standalone calculator page.
One-time setup
Monthly run-rate
Origin Lab
Project / local analysis
AI-likelihood signal check.
No upload
No API
No cost
Paste text or inspect an image in your browser.
Origin Lab gives a practical signal score, not a verdict. Text checks rhythm and repetition. Image checks metadata and pixel patterns. Nothing leaves this page.
First-pass inspection only. Never treat the score as proof.
Choose an image to inspect metadata and pixel signals.
--
Waiting for a sample
Local results will appear here.
No server request is made for analysis.
The browser does the work with ordinary JavaScript.
Use the score as a lead, not a final judgment.
Archive
About / contact
This is Malcolm's corner of the web.
MJGIVAI is a place for technical thinking, personal experiments, hobby notes, and durable links. It is intentionally small, static, and easy to change.