07 · The engine (your loop architecture, storified)

Three loops. One crew. You stay the captain.

Four steps handle every email. Three loops keep them running.

What happens to every email — step by step

✦ Connect
One harbor

Every account, any provider, behind one URL.

✦ Read
Triage first

Every mail scored, filed, or surfaced — nothing slips.

✦ Act
In your voice

Drafts, replies, follow-ups — written like you wrote them.

✦ Gate
You approve

Holds at the gate, every action logged, one tap sends.

The result: email that runs itself — with you holding the wheel.

AI agents with self-improving loops.

Floki's AI email agents improve themselves — safely. Every loop is contained, every action logged, and every increase in autonomy is a decision only you can make.

Learning loop

Builds a private model of your voice and habits from your edits and actions — per mailbox, never shared across boxes. 14 edits absorbed this week.

82% approve-unchanged
updated 1h ago
your voice & habits feed execution

Execution loop

Runs each of your workflows as mail arrives — trigger, AI steps, actions. This morning it filed 6 emails and drafted 2 replies, all within the trust dial you set.

214 handled / wk
ran 2m ago
patterns in the log feed improvement

Improvement loop

Watches outcomes and proposes new workflows or trust changes. Proposals only — you approve every one, and a dismissed idea stays dismissed.

2 proposals waiting
proposed yesterday

One continuous cycle: learning feeds execution, execution’s log feeds improvement — and approved proposals, plus your edits, flow back into learning. Proposals land in Suggestions; nothing enacts itself.