Personal agent + durable memory

Teach it once.
It remembers.

OSL is the chat agent you talk to. KnowShowGo is the memory brain underneath — facts, procedures, and private vault data that survive across sessions. You approve before anything sensitive submits.

Try the 3‑minute memory test Watch narrated demos

Cross-session recall
Private vs shared memory
Watch the live browser
Approval before submit

Two pieces. One product.

Cold visitors often mix these up — here’s the split in one glance.

OSL

Chat · Slack · Watch · form fill · vault ask-before-use

What you talk to

KnowShowGo

Concepts · assertions · embeddings · procedures · soft ACL

What remembers

Apps never call the memory HTTP API ad‑hoc — they use the KnowShowGo client SDK.

What memory actually does for you

Not “chat history.” Durable objects you can search, reuse, and keep private.

Teach once

Tell the agent your profile, logins (vaulted), and how a form should be filled. Next session it still knows — no re‑explaining.

Private vs commons

Your personal facts and credentials stay yours. Shareable procedures and ontology can compound in the public commons.

Watch mode

See the live Playwright browser while the agent works. When it’s unsure, it asks — it does not silently guess.

Forms that survive change

Procedures are remembered as data. When a layout shifts, the agent can repair and keep the fix — with your approval on submits.

See it in action

Narrated walkthroughs with captions and callouts — no dead air. Typed instructions in chat, live browser on the right, memory behind every fill.

Updated August 9, 2026 — real OSL Inbox (live IMAP) + Watch/Take-over on a live Greenhouse apply, narrated with audio.

Real OSL · Inbox + Watch (~40s)

No mock shells. The real /osl UI: ask “check my email” and the Inbox tab opens on live Gmail (IMAP) — Handshake codes, craigslist, the apply identity. Then real Watch footage of a Greenhouse apply with Take over always one click away, Gmail in the next tab, and Needs-you when a human must step in. Loud narration + captions.

Earlier montage (mixed sources)

Faster paced cut across Greenhouse chat demo, survey correction, selector heal, and Activity. Prefer the real Inbox+Watch clip above for posts.

Correction loop on mock forms (fixture)

Explicitly a local fixture — renamed-field Needs-you → correct → record → clinic intake. Not a live survey site. Use when explaining the repair story; use the real clip for social.

“Just reply to Dana” — and nothing else

One sentence in chat hands over one narrow slice of your inbox, and the assistant reads the limits straight back: her thread only, five a day, expires in thirty days. The Activity tab is the receipt — everything it may do without asking, everything it already did, and anything waiting on you. A different recruiter writes → it drafts and stops, because you never gave it her.

“Apply to Discord on Greenhouse” — then the next similar role

You type plain instructions in chat: remember my profile, apply to Discord’s Senior Software Engineer, Developer Experience role on Greenhouse. Watch fills the application from memory. You catch a missed LinkedIn field, correct it in chat, and the form updates. Then: apply to the next similar AI engineer role on Greenhouse — Grafana Labs, different field names, same facts, still stopped before submit. Narrated, captioned, with on-screen arrows.

“When Dana emails about the interview, just reply”

One sentence, and it handles the back-and-forth. A recruiter writes “can you do Thursday at 2pm Eastern?”; it works out the exact time, puts the interview on the calendar with a reminder an hour before, and answers her — because you said it could. Then a different recruiter writes, and it stops and waits for you. Permission is scoped to one person and one thread.

Teach once · fill two different ATS forms

The agent reads your résumé into KnowShowGo once, fills Employer A’s multi-step wizard, then reuses the same profile on Employer B — different labels and field ids — and stops before submit. That is the generalization story in one clip.

Automations that adapt

A site renames #login → #signin. The agent re-detects, heals the procedure, and saves the fix — so the next run survives the change.

KnowShowGo is the memory graph

Concepts, categories, and procedures as versioned data with provenance — not a prompt dump. Typed by prototypes; searchable by meaning.

Assertions with evidence and lineage
Embedding search over meaning
Soft private ACL (X-KSG-Owner)
Procedures resolved by intent, not hardcoded names
Reachable from your own code via the public API

Concept hierarchy browser (schema.org + WordNet seed)

Try it live — 3 minutes

Prove memory yourself. Don’t just watch a video.

Create a tester account /register → then open the agent
Teach one fact In chat: “Remember my favorite color is blue.”
New session, ask again “What’s my favorite color?” — it should recall from KnowShowGo.
Optional: Watch + Token Viewer /watch for the live browser · /tv/ to inspect stored objects

Teach it once.
It remembers.

Two pieces. One product.