The AI that actually knows you.
A profile is a long-form, narrative document your agent treats as canon. Health, faith, finance, fitness — each its own scope. Loaded by relevance, not by topic silo: your finance agent can read your health profile when you're pricing a procedure.
Beyond facts — a narrative the model can read.
A memory dump of “I am 36, I live in Durham, I am a staff engineer” is not who you are. Profiles construct a long‑form, structured story across identity, work, relationships, health, and faith — and every channel reads only what it needs.
Learn moreTradition
Catholic. Came in through RCIA, not cradle. Prefers a homily with teeth over sentiment.
Practice rhythm
Weekly mass at St. Thomas More; confession every six weeks. Re-reading Crossing the Threshold of Hope.
Primary condition
Ulcerative colitis, in remission. Humira biosimilar, biweekly, self‑administered Sunday nights.
Recent labs (Apr 2026)
Calprotectin 48, down from 180 at diagnosis. Next colonoscopy October.
Identity & rhythm
36, Durham‑based staff engineer, two kids under seven, lifts in the basement before the house wakes up. Daily rhythm built around the illness, not the ambition — gym, work, mass, repeat.
What I’m building
Patchwork — an incident‑response tool for small eng teams. The side project that might one day fund a four‑day week. Ships in TypeScript, ships small, ships often.
How to talk to me
Direct, technical, no padding. If something’s not working, say so before the fourth retry.
Default by domain, cross‑loaded by relevance.
Your health agent loads your Health profile by default. It can also pull in Financial when you’re weighing the cost of a procedure, or Spiritual when stress shows up in your sleep. Your life isn’t siloed by topic; your context shouldn’t be either.
Profiles update themselves.
When a thread surfaces durable context — a lab result, a salary change, a shift in conviction — the agent can commit a profile update. You get the diff, attribution, and restore path.
Versioned and inspectable.
Every profile is plain prose, diffable, and time‑stamped. Re‑take it any time, or roll back to last month’s you.
Profile templates, not one generic you.
A curated set of vertical templates ships out of the box — Personal (always loaded), Health, Faith, Fitness, OCD‑clinical. Each is a long‑form document the model treats as canon.
The taxonomy
Profiles fix the seven concrete ways frontier-chat memory fails. Each mechanic below addresses a specific failure mode.
Prose, not bullet facts.
Most AI memory is a flat list of one-liners: “User likes iced coffee.” “User uses tRPC.” “User wears Spicebomb Extreme.” A model reading that list has no idea which facts are load-bearing.
A profile is the opposite: long-form prose, organized into named sections (Genetics, Recent labs, Conditions, Medications…) that the model can read instead of guess at. You write it. You edit it. The agent treats it as canon.
Frontier chat
240 disconnected facts. Cologne preference next to calprotectin. No structure.
Boardbox profile
Markdown document with named sections. Versioned. Diffable. Auto-updated with restore.
Default by domain. Cross-loaded by relevance.
Each channel has a default profile it always loads —#health defaults to Health, #financedefaults to Finance. That’s the easy case.
The interesting case: when your conversation bridges, the agent loads additional profiles. Your finance channel can read Health when you’re weighing the cost of a procedure; your faith channel can read Personal when stress shows up in your sleep. We call this the drift principle — real life isn’t topic-siloed, so memory shouldn’t be either.
Every change is a diff. Versions persist.
Profiles ship with full version history. When the AI updates a profile — or you edit one by hand — the change lands as a new version with a diff against the previous. The version that was canon when an agent made a decision is the version we kept.
Diff
See exactly what changed and which thread or run caused the update.
Edit
Fix wording, soften confidence, add scope, or remove a stale project label.
Restore
Roll back to an earlier version without losing the audit trail.
The AI updates. You can audit and restore.
When a thread surfaces durable context, the agent can update the right profile directly. That is the point: memory should improve as work happens. The guardrail is that every update is visible, attributed, diffable, and restorable, so project-specific context does not silently become a universal fact forever.
Vertical profiles, not just a generic 'you'.
Personal workspaces ship with a curated set of vertical profile templates: Personal (the trunk), Health, Faith, Fitness, OCD-clinical. Each is a structured starting point you adapt — not a one-size-fits-all dump. The Personal profile is always frontloaded; the vertical profiles load on demand or by relevance.
Common questions
How is this different from ChatGPT memory?
Do profiles cross-reference between domains?
Where do I edit my profiles?
Can I see what changed when the AI updates a profile?
Are profiles private?
Related features
Indexes
Profiles are listed in the workspace index so the agent always knows which ones exist — and can load them on demand without you asking.
Learn moreChannels
Each channel has a default profile, plus the freedom to pull in others when the conversation calls for it.
Learn moreWorkspaces
Personal and Professional workspaces are isolated — your work agents never see your health profile, and vice versa.
Learn moreThe chatbox was a phase.
Build a workspace.
Boardbox keeps your context, holds its ground, and produces work that lasts.
Get started →