Find the workflowsworth automating.

Nolli audits the work you're already doing and turns it into instructions for AI Agents.The work already exists. Start using it as leverage.

Join the Founding Cohort

What we build

AI strategy derived from the work itself.

Using Nolli lets you leverage your everyday work as the roadmap for Agents to follow. The Agents just do what you do — your work is the guide.

Join the Founding Cohort
01

Nolli turns raw work activity into a readable workflow.

nolli audit 47 raw events
BROWSERopened workspace home
CALENDARreviewed today’s meetings
GMAILopened “Northstar — weekly update”01
BROWSERcopied account name
HUBSPOTreviewed account health + owner02
BROWSERswitched active tab
GMAILscanned previous thread
SLACKchecked #northstar-success03
SLACKopened pinned message
CALENDARreviewed client call notes
LINEARreviewed 3 open launch tasks
01Open client threadGmail
02Review account statusHubSpot
03Check unresolved issuesSlack
04Draft weekly updateDocs
Noise filtered
Reconstructed workflow

Weekly client status update

Pattern
01

Open client threadGmail

02

Review account statusHubSpot

03

Check unresolved issuesSlack

04

Draft weekly updateDocs

What we mean by “AI agent”

Agents are just folders
with a job

Nolli autonomously puts together the necessary folders & files needed to create a targeted agent for your work as it audits how the work gets done. Review at any time, or enrich the docs yourself.

instructions.md

An instructions.md file is basically the agent's operating manual. Tell the agent what to do, what not to do, and how to decide.

MD
instructions.md
Weekly client status agent
Assembled from audited work

# Instructions

You are a customer-support agent for Nolli.

## Goals
  • - Resolve customer questions quickly
  • - Use the knowledge base before guessing
  • - Escalate billing disputes to a human
## Behavior
  • - Be concise and friendly
  • - Never invent product features
  • - Ask one clarifying question only when necessary
## Tool Use
  • - Search /docs before answering
  • - Use the CRM tool to look up account details
  • - Never issue refunds without approval
## Output Format
  • - Give the answer first
  • - Then include next steps when relevant

tools/ & skills/

tools/ give the agent actions it can take & skills/ teach the agent how to perform repeatable tasks well. Connect your applications with APIs, MCPs, or local scripts the agent can execute.

capability.bundleAssembled
tools/
GmailRead · Draft
HubSpotRead accounts
SlackDraft message
local-scriptRun approved
skills/
triage-support.md
draft-client-update.md
escalate-risk.md
Selected skill
chooses the right tools in order.
Write actions still require approval

knowledge/

Ground the agent with static reference documents, database schemas, and domain specs.

knowledge/3 grounded sources
Approved sources
product-docs/
support-policies.md
account-schema.json
How are billing disputes handled?
Grounded answer

Escalate billing disputes to a human. Never promise a refund before approval.

policy §4.2refunds.md2 examples
Answers stay tied to source material

expertise/

Carry the judgment, exceptions, and institutional knowledge the agent needs to handle the work like someone who knows it.

expertise/decision-patterns.mdHuman reviewed
Judgment library

How experienced operators handle exceptions

14 cases
01At-risk accountEscalate before drafting
02Conflicting statusAsk the account owner
03Billing disputeRequire human approval
Exceptions stay explicit. The agent never turns confidence into permission.

Why AI agents

The Reality for Most Teams

The Problems

Building Agents takes entire teams
Existing scattered & repetitive workflows
Tasks fall through the cracks
Employees & teams are overwhelmed
Underutilizing AI beyond basic chat

The Outcome with Nolli

Recurring patterns surfaced
Candidate workflows scoped
Deliverables and handoffs mapped
Draft agents built with approvals
Progress measured against real outcomes

Privacy first

Understand the work without monitoring the people.

Nolli maps permitted work, not private activity. The privacy boundary is part of the product, not a policy page you are expected to trust.

Visible and stoppable

Capture stays on-screen, easy to inspect, and one click from paused. Nolli never runs as a hidden watcher.

Raw work stays on your device

Work events, transcripts, and permitted screen text remain inside your local Nolli workspace.

Sensitive fields are removed first

Secrets, credentials, payment data, and protected fields are stripped at the capture boundary.

Private apps start off-limits

Personal messages, photos, calls, and other private zones emit nothing unless you explicitly allow them.

Only approved patterns can be shared

Teams can learn from derived workflow structure without exposing the raw activity underneath it.

No employee scorekeeping

Nolli maps how work moves. It does not rank people, log keystrokes, or grade whether somebody looked busy.

Features

Real workflows.
Real automation.

Used by:Founders & ownersStartup operatorsMac + Chrome1-50 person teams
CClient follow-upHigh recurrence
WWeekly updatesReady for review
LLead handoffsNeeds context ?
IInvoice follow-upNeeds judgment

Prioritize the workflows worth fixing.

Nolli ranks the repeated workflows in front of you by recurrence, context, and judgment.

New
In Progress
Closed

Keep operating work moving.

Turn the work already happening across your tools into an approval-ready operating rhythm.

Client Status Agentinstructions.md
Keep the update shorter. Flag missing owner notes before drafting.
Added to the agent’s instructions. Approval stays required.
Spec updated · review required

Tune the agent with the work.

Chat with an agent’s instructions, add context, and keep the calls that need judgment with your team.

Founding access

Founder-led access. Priced at cost.

The first cohort is for Mac and Chrome founders and operators at 1-50 person companies. Exact pricing stays private while we measure real onboarding and operating costs.

Explorer

Private beta

For a founder testing the fit

For Mac and Chrome founders who want to see what their own work reveals before committing to a guided engagement.

  • Early product access
  • Personal workflow map
  • Direct product feedback channel
  • No self-serve automation promise
Apply for beta access
Founding program

Founding partner

At cost

Limited, founder-led program

For operators and small teams willing to work directly with Nadi on one real workflow worth improving or automating.

  • Short fit call
  • Guided install and consent setup
  • 3-5 day observation period
  • Evidence-backed workflow review
  • One approval-gated automation
  • Before-and-after review
Apply to the cohort

Enterprise

Later

Windows program in development

For larger teams that need Windows coverage, formal security review, and a repeatable rollout motion.

  • Enterprise interest list
  • Windows release updates
  • Security and deployment discussion
  • No premature enterprise promise
Join the enterprise list
Questions

Straight answers.

The things every team asks before they let something watch how they work.

Still wondering? Ask the founder

No. nolli is overt and consented — you turn it on, you see a live indicator while it's watching, and you turn it off in one click. It captures how you work, not keystrokes or passwords, and it's there to build agents for you, never to monitor you.

Nowhere, by default. Capture and analysis run on your own machine — local app, local database. Secrets are redacted before any text could ever leave. Nothing syncs to a server unless you explicitly choose to, and even then it's the redacted, organized evidence, never raw screens.

You'll see your own workflow map within the first day or two of normal work. nolli holds back from recommending anything until it has real evidence — so early on it observes quietly, then starts surfacing where AI fits as the pattern becomes clear. No six-week discovery engagement.

It's grounded in what you actually did, not what a model guessed. Counts and structure are computed deterministically; the AI only interprets on top, and it abstains instead of inventing a label when the evidence is thin. Every suggestion traces back to the moments that produced it.

The opposite, first. nolli starts by making your team better with the AI they already have — sharper prompts, the right skills, the tools that help. Automation comes later, only for the repetitive work nobody wants, and only with your sign-off. Your people move to the work that needs them.

A macOS app plus a Chrome extension today. The app watches work across every app on your machine; the extension reads the semantic work you do on the web. Setup is a 30-second wizard — no CLI, no config.