LinkedIn outreach that actually reads the profile

Five AI agents research each prospect, write something only that person would get, and pass it through a compliance review and a hard-coded rate limiter before anything sends. You approve. Nothing sneaks out.

Start free — no card Download desktop app

Dry-run by default Runs locally or in the cloud Bring your own AI key v0.1.7

agent-run · outreach · claude-opus-5

Live

Reasoning trace

6 steps · 11.4s · $0.031

read_profile

VP Engineering · 340-person Series C fintech · posted about on-call burnout 4d ago

search_knowledge

Retrieved positioning: “cuts incident triage time”, 2 fintech case studies

score_lead

ICP fit 87 / 100 — headcount, stage, and stated pain all match

reasoning

The burnout post is the hook, not the product. Lead with the specific problem they named, one case study, no pitch. Under 280 chars for a connection note.

Proposed action

connect+ note · 231 chars

“Your post on on-call burnout hit home — the 3am triage tax is what we spend all day on. We cut mean triage time 64% for a Series C fintech with a similar rota. Happy to share what worked, no pitch.”

Gates passed

Compliance agent review

no unverifiable claims

Policy engine · 24h caps

18 / 25 connects used

Working hours window

09:00–17:30 CET

Blocklist + dedupe

clean

Queued for your approval — nothing sends until you click

A real agent run, start to finish. Every trace is stored and replayable — you can always see why a message said what it said.

Mail-merge stopped working

Everyone on LinkedIn can spot a template now. The fix isn't a better template — it's not sending one.

Every other automation tool

One message, 500 recipients, a first name swapped in
You write the copy; the tool just clicks buttons
Limits you can raise until your account gets restricted
Replies pile up in an inbox you have to answer by hand
No record of why anything was sent

LinkedinAI

A fresh message per lead, reasoned from their actual profile
Agents write it; your Business Brain keeps them factual
Caps compiled into the code path — the AI cannot raise them
Replies get read, researched, and drafted for your approval
Full reasoning trace on every single action

Setup takes an afternoon

How it works

Connect your account

Sign in to LinkedIn once inside the app's own browser. On desktop, the session stays in a local profile on your disk. A warmup ramp starts new accounts slow.

Teach it your business

Point the Business Brain at your site, paste docs, or let it learn from past replies in your inbox. Every draft is grounded in what it retrieves — and nothing else.

Build a sequence

Collect profiles from search, Sales Navigator, or CSV. Then compose a multi-step workflow — visit, connect, follow up, pause on reply — on a visual canvas.

Approve and watch

Agents draft, the policy engine gates, you approve in the inbox. Read the full reasoning trace for every decision, then loosen autonomy once you trust it.

The agent team

Five agents, one narrow job each

Specialists beat a generalist. Each runs its own tool loop with its own system prompt, its own reasoning budget, and its own autonomy setting.

high effort

Prospecting

Sources and qualifies leads against your ICP — and scores a bad fit low, honestly, instead of forcing everyone through the funnel.

medium effort

Enrichment

Turns a raw profile into a 3–4 sentence brief and one concrete next action. Says what's missing rather than inventing it.

medium effort

Outreach

Drafts connection notes and openers in your brand voice, anchored to something specific about that person. Never a template.

medium effort

Reply

Reads the whole thread, looks the answer up in your Business Brain, and answers the question actually asked. Escalates what it shouldn't answer alone.

high effort

Compliance

Reviews every proposed action for spammy copy, unverifiable claims, and account risk. It can pause outreach; it can never raise a limit.

You set the leash

Suggest only — It proposes. Nothing is queued.

Draft & approve — You click send on every message.

Full auto — It runs. Guardrails still apply.

Per agent, per account. New workspaces default to draft & approve.

Everything included

The whole pipeline, one app

Sourcing to signed. No stitching four tools together with a Zap and hoping the data lines up.

Five specialist AI agents

Prospecting, enrichment, outreach, reply, and compliance agents running on Claude Opus 5 with a real tool-use loop — not one prompt wearing five hats.

Per-agent autonomy: suggest / draft / auto
Every step, token, and cost persisted
Prompt-injection–resistant tool design

Business Brain

A retrieval knowledge base of your positioning, pricing, objections, and voice. Agents may only state facts it returns — so they can't invent a feature you don't ship.

Scrape your site or upload docs
Learns from your past inbox replies
Chat with your brain to refine it

Profile collection & sourcing

Pull prospects from LinkedIn search and Sales Navigator result pages with paced pagination, or import lists you already own — de-duplicated on the way in.

CSV & Excel import
New-vs-seen counts as it runs
Workspace-wide never-contact blocklist

Visual workflow builder

Compose multi-step sequences on a canvas: profile visit, connection request, timed follow-ups, endorsements, InMail. Save them to a library and reuse.

Drag-and-drop step ordering
Spintax & custom variables
Auto-pause the sequence on reply

Campaigns that self-throttle

Run several campaigns per account with A/B tested variants. The engine ticks continuously, spreading actions across your working hours instead of bursting.

A/B variants with per-step stats
Smart throttling & randomised delays
Pause, resume, or kill instantly

Unified AI inbox

Every conversation across every connected account in one thread list, with an AI-drafted reply waiting. Approve, edit, or rewrite — then send in one click.

One-click draft approval
Bulk reply management
Auto-detected unsubscribe handling

Relationships CRM

A pipeline that fills itself. Leads move through stages as conversations progress, carrying their score, tags, deal value, and full message history.

Kanban pipeline stages
ICP scoring & custom tags
Complete per-lead conversation log

AutoPost content engine

Stay visible while you're outbounding. Generate posts on your themes, queue them on a calendar, and see which ones actually drove profile views and replies.

AI generator with campaign themes
Calendar queue & scheduling
Per-post performance analytics

Hard-wired guardrails

A deterministic policy engine sits between the agents and LinkedIn. Fifteen action types, each with rolling 24-hour caps the model has no tool to change.

Kill switch & account health checks
Working-hours windows per account
Seniority-based approval thresholds

Warmup & human pacing

New accounts ramp up over days rather than firing 100 invites on day one. Actions carry randomised delays and simulated keystroke timing.

Staged daily-limit ramp
Per-account persistent browser profile
Proxy routing configuration

Full audit trail

Every agent run, tool call, policy verdict, and executed action is written to an append-only log — readable in the UI, exportable when someone asks.

Step-level reasoning replay
Token spend per run
Dry-run mode logs without sending

Analytics & team

Funnel metrics per campaign, step, and account — acceptance, reply, and meeting rates. Invite teammates with scoped roles and manage seats per workspace.

Per-step funnel conversion
Team seats & role permissions
Bring-your-own AI keys

On the roadmap, not shipped yet

HubSpot, Salesforce, Pipedrive, Slack, Zapier and outbound webhooks are built into the UI but not yet live. We'd rather label them honestly than let you buy for them.

Coming in V1

Account safety

The AI can't talk its way past code

Guardrails written as prompt instructions are suggestions. Ours are a deterministic policy engine that runs between the agent and the browser — the model has no tool that writes to it.

Rolling 24h caps

Per action type, per account. Checked at execution time, not at queue time — so a backlog can't stampede.

Working hours

Actions only fire inside the window you set, in the account's own timezone. Nothing at 3am.

Kill switch

One toggle halts every account, every campaign, immediately. It's checked before each action.

Approval thresholds

Messages to senior titles route to a human for review regardless of the agent's autonomy level.

Dry-run default

Fresh installs simulate and log everything. Live sending is an explicit, deliberate opt-in.

Untrusted input

Scraped profile text is handled as data, never as instructions. A bio can't hijack an agent.

Read this before you buy

Automating LinkedIn violates LinkedIn's User Agreement. No tool, including this one, can make that risk zero — accounts can be restricted or banned. Everything above exists to make the risk small and visible, not to pretend it away. Live sending ships disabled; enabling it is your decision, made with the facts.

Your machine or ours

Identical agents, identical guardrails. The only question is where the browser runs.

Desktop app

Nothing leaves your disk

LinkedIn session stored in a local browser profile
SQLite database on your own filesystem
Chromium bundled — no setup, works offline
Bring your own Anthropic, OpenAI, or local model key
Only runs while your computer is on

Download now

Cloud app

Runs while you sleep

Dedicated cloud browser per LinkedIn account
Campaigns tick 24/7 — laptop closed, still sending
Unified inbox across every connected account
Team seats, roles, and sub-workspaces
Nothing to install; open it anywhere

Start free trial

Free download

Get the desktop app

Runs the browser automation locally on your own machine. Your LinkedIn session cookie never leaves your computer — and it keeps working offline.

Windows

Windows 10 or 11 · 64-bit

Version: 0.1.7
Size: 282 MB
Released: 2026-07-26

Download

Unsigned build — Windows SmartScreen shows “More info → Run anyway” on first launch.

macOS

macOS 12+ · Apple Silicon only

No macOS build published yet — signing and notarization in progress. In the meantime the cloud app runs everything in your browser, on any OS.

Use the cloud app

Works on macOS today — nothing to install

Linux

x86_64 · AppImage

Version: 0.1.7
Size: 359 MB
Released: 2026-07-26

Download

Run `chmod +x` on the AppImage, then launch it — no install step.

Dry-run by default

First launch simulates every action and logs it. Nothing is clicked until you explicitly turn live mode on.

Chromium bundled

No Playwright install, no Node, no terminal. The installer ships its own browser runtime.

Silent auto-update

Checks on launch and every 4 hours, then updates in the background. You never re-download by hand.

Simple pricing

Choose your tier

Same agents. Same guardrails. Different execution model.

Desktop

Privacy-first

$49/month

1 LinkedIn account · runs on your machine

Session never leaves your computer
All five agents (Claude Opus 5)
Human pacing & keystroke simulation
Warmup ramp for new accounts
Hard-coded compliance guardrails
Step-level audit trail
Local SQLite — works offline
Bring your own AI key

Download app

Windows & Linux today · macOS soon

Questions worth asking

Is automating LinkedIn against their Terms of Service?

Yes. LinkedIn's User Agreement prohibits automated access, and using any automation tool — this one included — puts your account at risk of restriction or permanent ban. We won't pretend otherwise. What we do is minimise that risk: conservative hard-coded action caps, human-like pacing with randomised delays, working-hours windows, a warmup ramp for new accounts, and per-account browser profiles. Live sending is off until you turn it on, and you accept that risk knowingly. If your LinkedIn account is business-critical and irreplaceable, the honest advice is not to automate it.

How is this different from Dripify, Expandi, or Linked Helper?

Those tools swap variables into a template you wrote — {{firstName}} into a message that's otherwise identical for all 500 leads. LinkedinAI runs a Claude Opus agent per lead that reads the profile, retrieves your positioning from your Business Brain, scores fit against your ICP, and writes something specific to that person. Then a separate compliance agent reviews the draft before anything is queued. Every step — the reasoning, the tool calls, the tokens, the cost — is persisted and readable in the UI.

Will the AI send messages without me seeing them?

Only if you explicitly choose that. Each agent has three autonomy levels: suggest-only (it proposes, nothing is queued), draft-and-approve (you review every message in the inbox and approve with one click), and full-auto. New workspaces start in draft-and-approve. You can also set a seniority threshold so messages to VP-and-above titles always route to you for approval, no matter the autonomy level.

Desktop or cloud — which should I pick?

Desktop if privacy is the priority: the browser, the session cookie, and the SQLite database all stay on your machine, and nothing runs when your laptop is closed. Cloud if uptime is the priority: a dedicated browser per account runs 24/7 on our infrastructure, campaigns keep executing overnight, and you manage multiple accounts from a unified inbox. Same agents, same model, same guardrails in both.

Do I need my own Anthropic API key?

On cloud plans, no — model usage is included. On desktop you can either use the included allowance or bring your own key in Settings → AI Models, which routes inference through your account at cost. The desktop app also accepts OpenAI and custom/local OpenAI-compatible endpoints if you'd rather not use Anthropic.

What stops the AI from raising its own sending limits?

Architecture, not instructions. The policy engine is deterministic code that runs after an agent proposes an action and before the executor can touch it: kill switch, account health, working hours, rolling 24-hour caps per action type, and approval thresholds. Agents have no tool that writes to it. A prompt-injected profile bio can't talk its way past code that never asks the model for permission.

Can I import my existing lead lists?

Yes — CSV and Excel import, with automatic de-duplication against everyone already in your workspace, plus a workspace-wide never-contact blocklist that survives imports. You can also collect profiles directly from LinkedIn search and Sales Navigator result pages, paginated and rate-paced, with new-vs-already-seen counts as it runs.

What happens to my data if I cancel?

Desktop data is a SQLite file on your own disk — cancelling changes nothing about your access to it. For cloud workspaces you can export leads, conversations, and the audit log at any time, and request full deletion, which we action within 30 days.

Stop sending mail-merge

Set up a workspace in an afternoon. Run it in dry-run until you trust what it writes — then let it go live on your terms.

Create free account Download for desktop

14-day trial on cloud plans · no credit card · cancel any time

Part of the iAgentice ecosystem. One sign-in carries a shared session across Card Studio, Calendar Sync, and Autopilot workflow nodes — hop between tools without logging in again.