Every step an agent takes is on a timeline. Who, what, when, and findable.

Every desk keeps a timeline: what it read, what it wrote, in which system, at what time, and every question a person answered, with their name on it. Nothing is edited after the fact. Open any run and you see every step: what it read, what it wrote, what it cost. Your records stay in your systems; we hold the process. Apps you run with us keep their data in a database of your own. This page is what your IT will ask for.

SCYTIAN · FINANCE · TIMELINE · WED 4 FEB

{{ r.who }} {{ r.did }} {{ r.obj }} {{ r.where }}{{ r.meta }}

WHAT THE TIMELINE ANSWERS

INVOICE 2026-0412 · BRATEX · FULL TRAILfound in 0.3 s

every desk has one · nothing is deleted · people’s decisions are on it too · kept for as long as you are a customer{{ foot }}

WHERE THINGS LIVE

Three places, each with its own job.

The agent reads a record, acts on it, writes the result back and keeps the log of having done so. Only apps you run with us keep data with us, in a database of your own.

ACCESS

The narrowest access that does the job. Written down, revocable in one click.

Read-only where reading is enough. One mailbox, not the domain. The bank is read, never paid from. Every connection names who granted it and when, and your IT sees the full list in the app at any time, without asking us.

SCYTIAN · CONNECTIONS · YOUR COMPANY7 connections · 0 admin rights

SYSTEMACCESSSCOPEGRANTED BYREVOKE

{{ c.sys }} {{ c.access }} {{ c.scope }} {{ c.by }} Revoke

revoking a connection stops the jobs that use it, nothing else breaksan agent never sees a key · it sees a door

KEYS

Pasted once. Encrypted on save. Never shown back.

API keys and passwords are encrypted the moment you save them. By default the value is never handed to the agent: an agent calls a connection by name and the platform opens the door, only for the system the key belongs to. The key itself never appears in a prompt, a log or a chat.

CONNECTION · XEROsaved 3 Feb · Anna K.

API key••••••••••••••••••Replace

Encrypted at restTLS in transitnever in a promptnever in a log

THE MODELS

The model sees the page it is reading. It keeps nothing.

Where a job needs an agent, the document or record goes to Anthropic’s Claude models under a business agreement: no training on your data, and retention only as those terms allow. A job that runs as a script sends nothing to a model at all, and most runs are scripts.

WHAT AN AGENT NEVER DOES

Seven things that never happen.

Everything else an agent does is written in a plan you approved and can read. These seven are in no plan we build, and the plan is the only thing an agent may act on.

FOR YOUR IT · THE SHORT VERSION

The facts, one line each.

Send this section to whoever signs off on vendors. The long version, with the data processing agreement, sub-processors and incident process, is one e-mail away.

Request the security pack

YOU CAN LEAVE

Your data was never here. The process leaves with you.

On the day you stop, you get every plan in plain language, every connection listed, and the full run log. Your container is deleted within 30 days and we confirm it in writing.

Bring your IT to the first call.

Thirty minutes, their questions first. The security pack is sent on request.

Book the call