Chatwoot

What it is

Chatwoot is an open-source omnichannel helpdesk: one shared inbox for website live chat, email, WhatsApp, Instagram, Facebook, Telegram, LINE, SMS and voice, with a help center, canned responses, automation rules, reporting and a public API around it. You can run it on Chatwoot’s cloud or on your own servers from the GitHub repository (37,000+ stars, Ruby on Rails and Vue, one-click Docker, Heroku and DigitalOcean deploys). Releases ship roughly monthly; v4.18.0 landed on 18 September 2026.

The AI layer is Captain. The Captain assistant answers customers first from your help center, FAQs and past conversations, then hands off to a human; Copilot drafts, rephrases, summarises and translates replies for agents. Captain assistants can also call up to 50 custom HTTP tools each (order lookups, refunds, account checks); the August and September 2026 releases added audience, schedule, scenario and outcome controls over when the bot speaks and what it may do. Captain runs on OpenAI models.

One thing to understand before anything else: Chatwoot is open core, not fully open source. Everything outside the repository’s enterprise/ directory is MIT-licensed. Captain, SLA policies, SAML SSO, custom roles, agent capacity limits, call transcription and audit logs sit under a separate enterprise license that requires a paid subscription in production.

Why it shows up on the shortlist

• It is the only credible self-hosted answer. Zendesk, Intercom, Freshdesk and Help Scout are SaaS-only. If your constraint is data residency, an air-gapped network or a regulator who wants conversations on your own infrastructure, Chatwoot is the default candidate.
• The per-seat price is a fraction of the incumbents’. A 10-agent team on Chatwoot Business pays $390 a month; the same team on Zendesk Suite Team pays $550 before any AI resolutions. Self-hosted Community costs $0 in licences.
• AI is priced per message, not per resolution. A Captain credit costs $0.02 at the $20-per-1,000 top-up rate, and one AI reply is one credit. Intercom’s Fin bills about $0.99 per resolved conversation. At 2,000 bot-handled conversations a month averaging three AI replies each, Captain costs roughly $110 in overage on the Business plan; Fin resolving half of them costs about $990.
• Migration paths exist. Recent releases added importers for Intercom (contacts and conversations) and Freshdesk (contacts, tickets, replies and private notes).

Pricing reality

Cloud, per agent per month: Hacker $0 (2 agents, 500 conversations, live chat only, 30-day retention); Startups $19 (all channels, help center, 300 Captain credits); Business $39 (adds voice, teams, automation rules, 500 credits); Enterprise $99 (SSO, audit logs, 800 credits, account manager from 20 agents). Third-party pricing analyses read the credit allowance as per account, shared by every agent, not per seat — so a 10-agent Business team gets 500 AI replies a month, not 5,000. Plan on buying credits.

Self-hosted: Community $0 (no Captain, voice, SSO, roles or SLAs); Premium $19 per agent (Captain, voice, branding, capacity management, email support); Enterprise $99 per agent (adds SAML and SLA policies). The real-world self-hosted cost is the licence plus a server with at least 4 GB RAM and 2 vCPUs, Postgres and Redis backups, and an engineer who applies security releases. For a team of 5 to 25 agents, budget $1,100 to $12,000 a year on cloud Startups or Business, before credits.

Best for

Heads of support and CX ops at SMBs and mid-market companies running 3 to 50 agents on chat- and WhatsApp-heavy volume, who want one inbox across messaging channels and are unwilling to pay Zendesk or Intercom seat prices — or who need the helpdesk on their own infrastructure and have an engineer to run it. Skip it if you need a mature AI agent that already resolves most tickets, deep ticket-form workflows with approvals, or an enterprise vendor behind a 500-agent contact centre.

Versus the alternatives

By installed base, Zendesk and Intercom own this budget. Pick Zendesk when you need structured ticketing, SLAs and a deep app marketplace; pick Intercom when AI resolution rate is the goal and Fin’s per-resolution price is cheaper than the agents it replaces. Freshdesk and Help Scout are the lower-cost SaaS picks when self-hosting is irrelevant. The fastest-growing entrants, Decagon and Sierra, sell autonomous agents that sit on top of an existing helpdesk — they complement Chatwoot rather than replace it. Zammad (AGPL, ticket-first) is the other open-source option; pick it over Chatwoot when email ticketing, not messaging, is your main channel. Chatwoot wins when you want messaging-first, self-hostable, and cheap per seat.

Watch-outs

• “Free and self-hosted” does not include the AI. Captain, SLAs, SSO and custom roles are enterprise-licensed. Guard: price the self-hosted Premium or Enterprise tier into the business case from day one, and list which of your must-haves live in enterprise/ before you pick Community.
• Captain has no published resolution benchmark. Its outcome, scenario and tool controls only shipped in August and September 2026. Guard: switch Captain on for one inbox, track resolution versus handoff with the conversation-outcome reports for 30 days, and scale only after it clears your target.
• Self-hosting AI still sends text to OpenAI. Captain calls OpenAI models, so conversation content leaves your servers even on your own install. Guard: get the processor terms reviewed, and keep Captain off inboxes that carry regulated data.
• You own the patching. Each of the last four releases carried security hardening, for sign-in, webhooks, uploads and more. Guard: name an owner, subscribe to GitHub releases, and apply each security release within a fixed window, such as seven days.
• The vendor is small. Chatwoot’s only disclosed funding is a $1.6M seed round in 2021, backed by Y Combinator. Guard: the MIT core means you can keep running a fork; also export contacts and conversations through the API on a schedule.