Open-Source Customer Support: Self-Host vs SaaS Break-Even

Open-source customer support software is real software. Chatwoot, Papercups, Libredesk - these are not toys. Teams use them in production. But 'free to download' and 'free to run' are different numbers, and for a 1-10 person team without a dedicated infra engineer, the gap between those two numbers is where the savings disappear.

The short version

If you have a dedicated engineer who enjoys infrastructure, self-hosting can work out cheaper at scale. If you do not, the hidden costs eat the savings inside 3-6 months. This post gives you the actual numbers so you can make the call without guessing. I built Trigli, a hosted SaaS that competes with these tools, so I have a stake in how you decide. I am going to try to be honest anyway.

Why small teams are looking at open-source right now

The Hacker News threads on Chatwoot and Libredesk are not astroturf. There is genuine interest, and it is not hard to understand why. Zendesk's pricing has crept up significantly - a 5-person team can easily find themselves looking at several hundred dollars a month before add-ons. Intercom layers per-resolution fees on their AI tier on top of seat fees, though the exact structure shifts often enough that you should check their current pricing page before assuming. The sticker shock is real.

The open-source pitch is compelling on paper: own your data, no per-seat fees, customize anything, no vendor lock-in. For teams that have been burned by a SaaS price increase mid-contract, that pitch lands. Worth understanding what is driving the interest before you either dismiss it or buy in completely.

What the main open-source options actually are

Chatwoot is probably the most mature of the three. Multi-channel (email, chat, social), active GitHub repo, real community, and they also offer a cloud-hosted tier if you want the software without the ops work. It is a legitimate product.

Papercups is lighter and more developer-focused. It is well-suited for embedding chat into a product. If you are a developer who wants something lean and hackable, Papercups is worth a look. Less feature-complete than Chatwoot but that is partly the point.

Libredesk is newer and has been getting attention on HN in the last few months, particularly for email-focused workflows. Still early, so verify current status before committing to it for production use.

What all three share: solid core features, no AI reply generation out of the box, and self-hosted by default. Worth noting that Chatwoot and Papercups both have cloud tiers, which changes the math significantly. A cloud-hosted open-source tool is basically just another SaaS with a different pricing model. The self-hosting conversation only applies if you are actually running it yourself.

The real cost of self-hosting: a line-item breakdown

Here is where most 'self-hosting is free' arguments fall apart. Let us go line by line.

Server costs: a properly spec'd VPS for a production support tool - with enough RAM to run Postgres, Redis, and the app itself without falling over during a spike - runs $20-60/month minimum. Add a second instance for redundancy and you are at $40-120/month before you touch anything else.

Maintenance time: OS patches, dependency updates, SSL renewals, backup jobs, log rotation. Estimate 2-4 hours per month conservatively. That is the floor, not the average.

Setup time: the first deploy is rarely under a half-day even for experienced engineers. Docker Compose or Kubernetes, environment variables, database setup, SMTP config, SSL, DNS. Call it 4-8 hours for a clean first deploy.

Opportunity cost: that engineer hour is worth something. Even at a conservative $75-100/hour internal rate, 4 hours of setup plus 2 hours of first-month maintenance is $450-600 in month one alone. That is before you touch anything else.

No AI included: Chatwoot, Papercups, and Libredesk do not ship with AI reply generation as a built-in feature. If you want AI drafts, you are wiring up your own LLM integration, building a knowledge base pipeline, and maintaining prompt logic. That is an engineering project, not a config toggle.

Downtime risk: if your support tool goes down at 2am, someone has to fix it. That someone is you, or whoever you can wake up.

Put it together honestly: a realistic self-hosted cost for a small team is $200-400/month in real dollars once you count engineer time at any reasonable rate. That is before you add AI. The $0 license fee is real. The $0 total cost is not.

Where Chatwoot and open-source genuinely win

Be honest with yourself about whether any of these actually apply to your situation.

You have engineering bandwidth and someone who actually wants to own this infrastructure. Not someone who will tolerate it - someone who finds it interesting. Chatwoot in particular has a real community and active repo, so there is genuine support when things break.

You are at a scale where per-seat SaaS fees become genuinely painful. That usually means 20+ agents. Below that, the math rarely flips.

You have data residency requirements that make third-party hosting a compliance problem. Healthcare, certain financial services, some EU contracts. Real constraint, and self-hosting is often the only clean answer.

You want deep customization that no SaaS will ever offer. Custom workflows, custom data models, integrations into internal systems that no vendor will build for you. Chatwoot's open codebase makes this possible in a way that hosted tools simply do not.

You are building a product on top of it, not just using it for support. Embedding a support layer into your own product is a different use case than running support for your team. Papercups was built with exactly this in mind.

If one of those is genuinely true for your team, open-source is a serious option and you should pursue it. If you are reaching to make one of them fit, that is a sign you are rationalizing the free license fee.

The break-even math for a 1-10 person team

Three scenarios worth running through.

Scenario A: founder or one engineer handles support, roughly 200-500 emails and chats per month. At that volume, a hosted tool at $49-149/month is almost certainly cheaper than the engineering hours to self-host. The Trigli Growth plan at $149/month covers 2,500 emails, 1,000 chats, and 500 tickets. You are not going to beat that with a VPS and your own time. See the full breakdown at /pricing.

Scenario B: team of 5-8, 1,000-3,000 tickets per month, no dedicated infra engineer. Self-hosting adds real operational risk with no clear cost win. The per-seat fees on most SaaS tools at this team size are still lower than the engineering hours to maintain a self-hosted stack.

Scenario C: 10+ person team with a DevOps person already on payroll who has spare cycles and wants the project. Now the math starts to flip. The self-hosting overhead gets absorbed into an existing salary, and per-seat SaaS fees start to sting at 15-20 agents.

The break-even point for most small teams with no dedicated infra person is roughly 15-20 agents where per-seat fees exceed $300-400/month. Below that number, hosted SaaS wins on total cost of ownership almost every time. The middle ground - 10-15 agents, one person who could handle infra but has other priorities - is genuinely a judgment call.

What you give up on AI with self-hosted tools

This is the part most comparisons skip. Chatwoot, Papercups, and Libredesk do not ship with AI reply generation as a built-in feature. You can integrate OpenAI or another LLM, but that is an engineering project. You need to build the knowledge base ingestion pipeline, decide on confidence thresholds, build the draft-vs-auto-send logic, and then maintain all of it when the API changes or your docs update.

For a team that just wants support handled, this is a significant gap. The AI piece is often the whole reason someone is looking at a new support tool in the first place. If you are reading this because you want to stop drowning in repetitive email, self-hosting a tool that does not include AI and then building your own AI layer is not a shortcut. It is a longer path.

Trigli does this out of the box: reads from your docs only (PDFs, pasted text, URL imports), flags uncertain answers for human review rather than guessing, and learns tone from your past sent emails before drafting replies. That is not a trivial amount of work to replicate yourself on a self-hosted stack. The post at /blog/ai-email-autoresponders-small-business goes into more detail on what to look for in AI-bundled tools if that is the direction you are leaning.

The setup time comparison is not close

Self-hosted: Docker Compose or Kubernetes, environment variables, database setup, SMTP config, SSL, DNS. Call it 4-8 hours minimum for a clean first deploy. That is if everything goes right. It often does not.

Hosted SaaS: OAuth connect to Gmail or Outlook, paste in your docs, set your automation rules. Most teams are live in under an hour. Trigli connects via OAuth 2.0 so replies still send from your real email address - no second inbox to manage, no DNS changes. That is the actual difference between a tool you use on day one and a project you are still configuring on day three.

For a founder who just wants support handled, that delta matters. The /blog/cost-of-support-rep-vs-ai post has more on what the real cost of unhandled support volume looks like when you are the one answering every email.

Where Trigli fits in this decision

Trigli is a hosted SaaS. Flat pricing, no per-resolution fees, bundles email plus chat plus ticketing plus AI in one plan. The Growth plan at $149/month covers 2,500 emails, 1,000 chats, and 500 tickets. That is a real small team's full month of support volume.

It connects to Gmail and Outlook via OAuth 2.0. Both are live today, not on a roadmap. Replies send from your real address. The AI answers from your docs only and flags uncertain ones for human review rather than guessing. Built-in ticketing includes priority levels, assignment to team members, internal notes, and SLA tracking. Not a bolt-on. It is part of the same plan.

Automation rules let you route incoming messages by category to one of four actions: auto-send, draft, human review, or skip. Below the confidence threshold, nothing auto-sends. You control what goes out. The /use-cases/small-business page has more on how small teams are using it in practice.

What Trigli does not do

I would rather tell you now than have you find out after you have set things up.

No phone or SMS support. Email, chat, and tickets only. If your support volume is heavy on phone calls, Trigli is not the right fit.

No per-agent productivity dashboards. You cannot pull a report showing tickets resolved per agent per week. If that reporting matters to you, look at Zendesk or Freshdesk.

No skill-based agent routing. Routing is by ticket category to an action, not by agent expertise. If you need tickets auto-assigned to the agent with the right skill tag, that is not something Trigli does today.

No native Shopify actions. The AI can reference your refund policy and tell a customer what it says, but it cannot process a refund inside Shopify. That action still requires a human.

No multi-brand routing inside one account. One tenant equals one brand. If you are running two separate brands with different inboxes and different support teams, you would need two accounts.

If any of those are hard requirements, Trigli is not the right fit and there is no shame in that. The honest comparison at /blog/ai-email-autoresponders-small-business covers other tools that might fill those gaps.

How to actually make this decision

Ask yourself one question first: do I have an engineer who wants to own this infrastructure, or am I the engineer and I have other things to build? That single question resolves most of the decision.

First, count your monthly volume honestly. Emails, chats, tickets. Write the actual number down. Then price out the self-hosted total cost: server ($20-60/month) plus 2-4 hours of engineer time at your real hourly rate - not your billing rate, but what that time is actually worth to your business. Compare that to a hosted plan at that volume. Most small teams land between $49 and $149/month. If the numbers are close, factor in AI, setup time, and who fixes it at 2am. Most teams under 15 agents land on hosted. Most teams over 20 with infra capacity land on open-source. The middle is genuinely a judgment call and you should make it with your own numbers, not someone else's.

The honest summary

Open-source customer support software is not a scam. Chatwoot is used in production by real teams at real scale. The free license is genuinely free. What is not free is the time to run it, the time to maintain it, and the engineering work to add AI on top of a stack that does not include it.

For a 1-10 person team without a dedicated infra person, hosted SaaS wins on total cost of ownership almost every time. The break-even is roughly 15-20 agents, and most small teams are not there yet.

If you want to try the hosted route without committing, Trigli has a free tier (50 emails, 25 chats, 10 tickets) and a 14-day trial on paid plans - card is collected at signup but you are not charged until day 15 if you do not cancel. No hard feelings if self-hosting is the right call. If the numbers do not add up that way, worth a quick look at trigli.com.