OpenOutreach Setup Guide 2026: Free AI LinkedIn Outreach

Here's a scenario I hear constantly from founders: you know exactly who your customer is. You can describe them in one sentence. "VP of Engineering at Series B SaaS startups that are scaling their backend team." Perfect ICP. Crystal clear.

And then what? You open Apollo, buy a list of 500 contacts for $300, spend a weekend writing personalized messages, click send, and wait. Three months later: 12 replies, 2 demos booked, and a nagging feeling that you're burning money to send emails that feel like spam.

The problem isn't your targeting. It's the tooling. Most outreach tools were built around a fundamental assumption: you already have the list. They help you send to people. They don't help you find, qualify, and learn who the right people actually are.

OpenOutreach flips this completely. You tell it what you sell and who you're targeting. It finds the leads, scores them with AI, sends the messages, handles the follow-ups, and gets smarter with every response it receives. The whole thing is free, open-source, and runs on your own machine.

This guide will walk you through setting it up - step by step, in plain English, even if you've never used Docker or touched a terminal before.

"The best outreach list isn't one you buy. It's one your AI builds and refines over time."

Before we touch any settings, it's worth understanding what problem this tool was actually built to solve. Because once you get it, the setup decisions make a lot more sense.

The $200/month problem

LinkedIn automation tools - PhantomBuster, MeetAlfred, Waalaxy, Dux-Soup - are useful. But they're also expensive. At 2–5 paid seats, most teams spend $150-400/month just for the software. On top of that, you're still buying the lead lists from Apollo or ZoomInfo. And using their shared proxy infrastructure, which LinkedIn has gotten very good at detecting.

The founders behind OpenOutreach looked at this and asked a different question: what if the tool was free, ran on your own machine (so it uses your IP, not a shared proxy pool), and didn't require you to build the list first?

The three things that make it different

• →

  You describe your ideal customer in plain language. The AI writes LinkedIn search queries, runs them, and discovers candidates automatically. You never touch a spreadsheet.

• →

  Every time someone connects, replies, or ignores you, OpenOutreach updates its internal model. Over weeks, it becomes much better at predicting who's worth targeting - a feedback loop that static lists can never provide.

• →

  Everything lives in a local database file on your machine. No vendor has your leads, your conversations, or your results. GDPR-friendly by default.

What it actually does, in plain English

Imagine you hired a tireless researcher who:

1. Reads your product description and writes LinkedIn search queries
2. Finds hundreds of matching profiles and checks if they're actually a good fit
3. Sends them a personalized connection request
4. When they connect, starts a conversation - and adjusts based on what they say
5. Logs everything in a spreadsheet you can always see

That's OpenOutreach. The "researcher" is a combination of an LLM (like GPT-4) and a statistical model that learns from your results.

Setting up OpenOutreach takes about 30-60 minutes. Here's everything you need before you begin.

The essentials

• →

  Strongly recommended: create a dedicated secondary LinkedIn account for automation. Your main account is too valuable to risk. A fresh account with a realistic profile photo, a job history, and 50+ connections looks legitimate to LinkedIn's detection systems.

• →

  OpenOutreach uses an AI model to generate search queries and write messages. It works with OpenAI (GPT-4), Anthropic (Claude), or any OpenAI-compatible API. If you don't have one, go to platform.openai.com and grab an API key - you'll spend maybe $5-15/month at typical usage levels.

• →

  Docker is the easiest way to run OpenOutreach. Think of it as a special container that runs the software in an isolated environment, so you don't have to install anything complicated. It's free. Download it at docker.com/products/docker-desktop.

What you do NOT need

• Any technical background
• A lead list, CSV, or database of contacts
• A paid subscription to any other tool
• A server or cloud hosting (at first - that's optional later)

We'll use Docker - the fastest, most beginner-friendly path. It runs the entire OpenOutreach system in one command.

Now that the tool is running, let's make it actually do something useful.

Access the admin dashboard

Open a second browser tab and go to: http://localhost:8000/admin/

This is OpenOutreach's built-in CRM - a clean interface showing every lead in your pipeline, what stage they're at, and the results so far. The default login is username: admin and password: admin - you'll be prompted to change this on first login.

Define your campaign

In the admin dashboard, click Campaigns → Add Campaign. You'll fill in:

Product Description - Be specific. Not "we help companies with AI" but "we build AI agents that automate customer support ticket routing for SaaS companies with support teams of 10+ people." The more precise this is, the better the AI's targeting.

Target Market - Again, specificity wins. "VP of Customer Experience or Head of Support at B2B SaaS companies, Series A or later, 50–500 employees" will outperform "B2B companies."

Daily Connection Limit - Start conservative. Set this to 15-20 per day. LinkedIn flags accounts that send too many requests too fast. Lower is safer, especially on a newer account.

Message Template - OpenOutreach generates personalized messages automatically, but you can set the tone and include key points. Example:

Start the campaign

Once your campaign is configured, click Activate. OpenOutreach will:

1. Generate LinkedIn search queries based on your target market
2. Open LinkedIn in the automated browser and run the searches
3. Pull profile data for matching results
4. Score each profile (this is the AI qualification step)
5. Send connection requests to high-scoring profiles, paced over the day
6. When someone connects, begin a follow-up conversation

You can watch all of this happen in real time at localhost:6080.

This is the part that makes OpenOutreach genuinely different from every other tool on the market. It's worth understanding, even at a high level.

Most outreach tools use filters: industry = SaaS, title = VP, company size = 50–500. These filters are static. They don't improve. A VP of Engineering at a 200-person company matches the filter whether they're a great prospect or a terrible one.

OpenOutreach uses a different approach. After each connection request you send, the system notes whether the person:

• Ignored the request
• Connected but didn't reply
• Connected and engaged in conversation
• Became a real lead

Over time, it builds a statistical model: "People with these characteristics (seniority level, industry keywords, company growth stage, mutual connections, profile completeness) tend to engage. People with those characteristics don't."

This is why the first two weeks feel slower than expected - the model is learning. By week three or four, if you've run consistent volume, you'll see a meaningful improvement in connection acceptance rates and reply rates.

One thing to watch: AI learning loops can also drift. If the model starts over-indexing on a narrow profile type or the conversation agent starts generating off-brand messages, you want to catch it early. We wrote about how to spot and measure agent drift if you want to go deeper on this.

This section is important. LinkedIn's detection systems have gotten significantly better in 2025–2026. They look at behavioral patterns - not just the tool you're using. Here's how to stay well within safe limits.

• →

  Week 1: max 10 requests/day. Week 2: 15-20. Week 3+: up to 25-30 if engagement looks normal. Accounts that spike to 50+ requests on day one get flagged immediately.

• →

  A two-week-old LinkedIn account with 12 connections looks suspicious. Before running automation, spend 2–3 weeks building the account naturally: connect with people you actually know, post once or twice, join a few groups.

• →

  Long messages with multiple links, bullet points, and a sales pitch in the first message are a red flag - both for LinkedIn detection and for real humans. Keep opening messages under 300 characters. No links. No attachments.

• →

  Configure OpenOutreach to only run during business hours in your timezone. Real humans don't send LinkedIn messages at 3am. Run 9am-6pm, Monday through Friday only.

• →

  If your acceptance rate drops below 15%, stop. Either your profile isn't compelling or your target market isn't a match. Adjust before sending more volume.

OpenOutreach isn't the right tool for every situation. Here's an honest comparison.

Choose OpenOutreach if:

• You want zero ongoing software costs
• You're willing to spend one hour on setup
• You care about data ownership and privacy
• Your targeting is precise enough to describe in words
• You're running outreach over weeks or months (the AI gets better)

Choose a SaaS tool if:

• You need to be running in 30 minutes with no setup
• You're testing outreach for the first time and don't want infrastructure overhead
• Your team is non-technical and can't help if something breaks
• You're doing a short, one-off campaign and don't care about long-term learning

Running OpenOutreach on your laptop works fine - but your laptop needs to be on for it to run. If you want it running 24/7 without babysitting your machine, there's a managed cloud option.

From your terminal:

This installs the OpenOutreach CLI, which lets you upload your local database to a managed server. The same open-source code runs there - no feature gating, no data locked in. You can switch back to self-hosting at any time and your data comes with you.

This is worth doing once you've validated your campaign on your local machine and want to scale up. Don't start here - start local, get results first, then decide if the managed option makes sense for your volume.

After the first week, open your admin dashboard at localhost:8000/admin/ and check these numbers.

Connection acceptance rate

Target: 20%+. If you're below 15%, something's off - either your profile doesn't match your target market's expectations, or your targeting is too broad. Narrow your ICP description and look at which profiles are accepting vs. ignoring you.

Reply rate (among connected)

Target: 10–20%. This depends heavily on your message quality and the fit of your offer. If people are connecting but not replying, your opening message isn't creating enough curiosity. Make it shorter, more specific to their situation, and end with a question. The best way to improve this systematically is to run two message variants side by side and measure - we cover exactly how to A/B test AI agent configurations in a separate post.

Qualification score distribution

In the dashboard, look at the profile score histogram. Early on, it should be spread across a wide range. Over time, as the model learns, you should see more high-scoring profiles in your pipeline and fewer low-scoring ones being contacted. This shift is the Bayesian learning working.

"The Docker command isn't working" Make sure Docker Desktop is open and running (not just installed). The app needs to be actively running in the background before you can use Docker commands.

"I can't see anything at localhost:6080" Give it 30–60 seconds after the startup command - the browser environment takes a moment to initialize. If it still doesn't load, try restarting the Docker container by pressing Ctrl+C in the terminal and running the startup command again.

"LinkedIn keeps asking me to verify my account" This happens on newer accounts or after login from an unfamiliar location. Log into LinkedIn normally in your regular browser first, complete any verification it asks for, then restart OpenOutreach. The tool will use the same cookies once you're verified.

"Messages aren't sending" Check your LLM API key first - if it's invalid or out of credits, the message generation step will silently fail. Log into your OpenAI or Anthropic dashboard and confirm you have active credits.

"My data disappeared after restarting" This means the -v ~/.openoutreach/data:/app/data part of the startup command wasn't included. That flag mounts a persistent storage location. Add it to the command and restart - your future data will persist correctly.

Once you've run your first two weeks and have real data:

1. Review your top 20 accepted connections - what do they have in common that your target description didn't capture? Update your ICP to match.
2. Look at the conversations the AI is having - are they progressing naturally or stalling? Adjust your follow-up prompts. If you're running other AI agents in your stack, Syrin's agent traces give you step-by-step visibility into exactly what each agent decided and why.
3. Run message variants - don't guess which opening line converts better. Test two versions and let the data decide. Our post on A/B testing multi-agent systems walks through how to do this rigorously.
4. Increase volume gradually - once acceptance rates look healthy (20%+), bump your daily limit up by 5 connections per week.
5. Consider the managed cloud option - if you're closing deals and want 24/7 uptime without keeping your laptop on, the hosted version is worth evaluating.

The pattern that works: small batches, close observation, continuous adjustment. OpenOutreach's learning loop rewards patience. Teams that run it for 60+ days consistently see better results than teams that blast high volume in week one and burn out.

"Set it and forget it is a myth. The winning formula is: set it, watch it, improve it."

Continue reading

AI Experiments

Your AI Agent Has Been in Production for Weeks. Do You Even Know What's Working?

May 19, 2026Read

Research

Agent Drift is Now Quantifiable: Here's the Math Behind It

April 28, 2026Read

Observability

Why AI Agents Drift Away From Their Goals (And How to Stop It)

April 20, 2026Read