Local-first · Draft-only · Auditable

Your Chief Of Staff
helping you take action

Kermit reaches every system the orchestrator can see — mail, calendar, project trackers, CRM, chat, meeting notes, analytics, databases, code, the open web and beyond — and turns the noise into one clear, auditable briefing every morning.

systems reachable

anything the orchestrator can see — apps, data, web

1briefing each morning

ranked, source-linked, ready at 6:30 AM

0records ever modified

read-only everywhere; drafts only

Why Kermit

A clearer pass across every system you rely on.

Most mornings start with the same tax: open every tool you own, reconstruct what changed overnight, decide what actually matters. Kermit makes that pass across any system the orchestrator can reach — and shows its work.

Without Kermit

  • Context scattered across every system
  • Urgent items buried under noise
  • No audit trail for “why this mattered”
  • Manual triage before real work begins

With Kermit

  • One briefing, ranked by what matters
  • Escalations surfaced, never buried
  • Every item carries a source link and an epistemic tag
  • Drafts ready; you stay the decision-maker
What you get

Three things, done right.

Morning clarity

A single daily briefing — schedule, conflicts, meetings needing prep, urgent actions, drafts, and cross-source themes — ready before your day starts.

Decisions, not data dumps

Every item is classified — Decision, Action, Watchlist, Draft, Escalation, System Pattern or Win — with owner, next step and urgency.

Trust by construction

Read-only everywhere. Draft-only output. Each claim tagged [FACT] / [INFERENCE] / [ASSUMPTION] / [NEEDS REVIEW] and linked to its source.

How it works

A four-stage pipeline, every morning.

Kermit observes approved sources, reasons over them through two lenses, classifies what matters, then writes the briefing and dashboard in one atomic, auditable operation.

01

Observe

Any reachable system: mail, calendar, chat, trackers, CRM, tickets, docs, meeting notes, analytics, databases, code, APIs, the open web.

02

Preflight

Verify each source: OK / PARTIAL / MISSING. Gaps are flagged, never guessed. Fail-loud on access errors.

03

Reason

Two lenses: a Priority Lens that weights by reliability, throughput, adoption, usability and security; and a Pattern Lens that classifies recurring signals.

04

Classify + Write

Seven item classes, each with owner · next step · urgency · source · confidence. Briefing and dashboard written in one atomic operation.

Kermit system architecture: connected read-only sources feed an engine-agnostic reasoning core; outputs persist to a local-first vault; a governance rail enforces draft-only, read-only, and auditable guarantees.
Kermit's system architecture — connectors in, reasoning core, local-first vault, governance rail.
Reach

Any system the orchestrator can reach.

Kermit isn't limited to a fixed list of integrations. It can observe anything its orchestrator can see — and the set keeps growing.

Mail
Calendar
Chat
Project trackers
CRM
Tickets
Docs
Meeting notes
Analytics
Databases
Code & APIs
The open web
…and whatever the orchestrator connects next.
Built for trust

The guardrails that make Kermit safe to point at real systems.

Six hard rules, enforced by construction — and the architecture that keeps you in control.

R1

Draft-only

Never sends, posts, or submits. Output is always a draft for you.

R2

Read-only external

No record in any connected system is ever modified.

R3

No secrets in memory

Credentials and sensitive data are never persisted.

R4

Epistemic tagging

Every claim is labeled FACT / INFERENCE / ASSUMPTION / REVIEW.

R5

Fail-loud

Missing sources are flagged by preflight, never silently skipped.

R6

Work / personal split

A strict boundary; no personal data in work artifacts.

Under the hood

Local-first

Markdown is the source of truth in your own Kermit/ vault; your data stays on your machine.

Multi-engine

Pluggable inference engines behind one contract — swap models without changing the system.

Atomic deploy

Briefing, run log, reasoning trace and dashboard render in a single approved write.

Auditable runs

Every run leaves a reasoning trace and run log — replayable and reviewable.

What you see

One briefing. One dashboard. Fully sourced.

Each morning Kermit produces a structured briefing (the source of truth) and a dashboard view with team tabs, priority badges, and deep links straight to the underlying record. Feedback chips let you correct or confirm — tuning future runs without ever changing a live system.

[FACT][INFERENCE][ASSUMPTION][NEEDS REVIEW]
Morning briefing· Today · 06:30 PT
Kermit/60-dashboard/
  • P1Decision[FACT]
    Renewal terms need your call
    due Fri · CRM · Owner: you · [source]
  • P1Escalation[FACT]
    Customer thread waiting 36h
    Mail · ref #2418 · [source]
  • P2Action[INFERENCE]
    Draft reply to investor update
    ready for review · Drafts · [source]
  • P2Watchlist[INFERENCE]
    API error rate trending up
    Analytics · last 24h · [source]
  • P3System Pattern[ASSUMPTION]
    Recurring boundary confusion across teams
    Pattern Lens · [source]
  • P3Draft[NEEDS REVIEW]
    Weekly ops summary prepared
    Briefing · 6 sources · [source]
7 team tabs · 6 sources · 1 reasoning tracerun-2026-06-17.md
FAQ

Common questions.

No. Kermit is draft-only and read-only by design. It prepares drafts and surfaces what matters; you decide and act.

Locally. Markdown is the source of truth in your own Kermit/ vault on your machine.

Anything the orchestrator can reach — mail, calendar, chat, trackers, CRM, analytics, databases, code, the open web, and more over time.

Every item carries an epistemic tag and a link to its source, plus a reasoning trace for the whole run.

No. Kermit is multi-engine — inference engines are pluggable behind one contract.

See a morning briefing in action.

Kermit explores safe, auditable agentic operations — reaching any system the orchestrator can see, drafting never sending.

mykermit.aiA Blank Page project