Mike HeapMike is an experienced Product Manager who focuses on all the “non-development” areas of My AskAI, from finance and customer success to product design, copywriting, testing and more.
Table of Contents
- OpenCX at a glance
- What is OpenCX?
- How easy is it to set up OpenCX?
- What channels does OpenCX work in?
- What are the limitations of OpenCX?
- What knowledge sources can I train OpenCX on?
- What features does OpenCX have?
- The autonomous agent
- Actions and workflows
- Agent assist
- Outbound
- How do I improve OpenCX responses?
- What resolution rate can I expect from OpenCX?
- What AI model does OpenCX use?
- What languages does OpenCX work in?
- How secure is OpenCX?
- Who is using OpenCX?
- How much does OpenCX cost?
- What OpenCX charges for
- What counts as a billable outcome
- The $2M guarantee
- What it costs at 2,000 conversations a month
- Does OpenCX have a free trial?
- Is OpenCX worth it?
- What are the Pros and Cons of OpenCX?
- Pros
- Cons
- FAQs
Created time
Sep 9, 2026 07:02 AM
Title length (<60)
Author
Last optimised
Ecomm?
Tags
Image
Publish date
Aug 17, 2026
Video
Slug
opencx-complete-guide-2026
Featured
Type
Article
Ready to Publish
Ready to Publish
OpenCX charges between 75 and 90 cents for every ticket its AI closes, and nothing when one reaches a human. At OpenCX's published 77% resolution rate, 2,000 conversations a month means 1,540 resolutions, and the bill lands between $1,155 and $1,386.
OpenCX is an AI support agent sold by OpenChat, Inc., the Y Combinator-backed company behind the open.cx platform. One trained agent picks up phone calls, web chat, email, WhatsApp, SMS, Slack and social posts, and it installs over the helpdesk you already own with no migration. Billing follows outcomes, which suits a high-volume team ready to negotiate a contract and gets in the way of one that wants to test first.
I'm guessing you sat through a demo, the product looked good, and then the price came back as a range.
Now you have to turn that range into a figure your finance lead will accept. You have a ticket volume, a rough sense of how much AI could take, and a helpdesk nobody wants to rip out.
I run an AI support agent business, so I spend a lot of time inside other people's pricing pages. By the end of this you will know what it costs at a real volume and what it counts as a resolution.
OpenCX at a glance
Fact | Detail |
What it is | AI support agent across every channel |
Headline price | $0.75 to $0.90 per resolved ticket |
Best for | High-volume teams running voice and text |
Not a fit for | Teams that need to trial before buying |
Biggest limitation | Rate and resolution definition are both negotiated |
Our verdict | 7.5/10 |
My AskAI bills for every ticket its AI works on, resolved or not, and publishes the rate for each plan so you can build the case before you speak to anyone. See our pricing explainer for the worked numbers.
What is OpenCX?
TL;DR: Six agents run off one engine: Inbound, Outbound, Assist, QA, Customer Understanding and Apps. Train once, and the whole thing installs over your existing stack with no migration.
The company's homepage names those six. Inbound is the customer-facing one, answering on every channel and handing over when it should. Assist sits with your agents, drafting replies and filling fields, while QA and Customer Understanding read the conversations afterwards.
I look at where OpenCX sits in your stack. You keep your helpdesk:
"Zendesk, Intercom, Salesforce, HubSpot, Freshdesk, Twilio Flex, or Open's own. It sits on top of the stack you already run. No migration."
The company brands itself Open, and the about page is where OpenChat, Inc. sets out that positioning. Y Combinator backs it, and OpenCX announced a $7M round led by YC and X By Unifonic, with Shorooq and Sadu taking part.
My AskAI takes the same position in the stack. Our agent installs into Zendesk, Intercom, Freshdesk, Freshchat, Gorgias or HubSpot, and if you switch helpdesk later the trained agent moves with you. See our helpdesk integration pages.
How easy is it to set up OpenCX?
TL;DR: Five steps to a live web agent, under ten minutes per OpenCX. Helpdesk, phone numbers and your own APIs take longer, and the commercial side runs through sales.
There is a self-serve way in. The OpenCX Quickstart opens by telling you to create an account (that signup page loads without a sales conversation). Its five steps promise a live agent on your site in under ten minutes:
- Create your account.
- Connect one knowledge source at Settings → Knowledge.
- Turn on Autopilot for the web channel at Settings → Autopilot.
- Paste the widget snippet from Channels → AI Chat Widget just above your site's closing body tag.
- Ask it a question and watch what it does.
A support lead can do all five without help, and step four is the only one that touches your website's code (a five-minute job for whoever manages your site).
The rest takes longer. Connecting your helpdesk, setting up phone numbers and wiring AI Actions into your systems are separate jobs. The helpdesks page promises you will be live inside your existing stack in under an hour, a fair claim for a connector.
You cannot shortcut the commercial side. The pricing page is clear that your rate is agreed before you go live, so a real deployment starts with sales.
My AskAI installs from your helpdesk's own app marketplace, and the standard setup takes about ten minutes with no developer. We wrote separately on how long a rollout actually takes.
What channels does OpenCX work in?
TL;DR: One agent covers voice, web chat, email, WhatsApp, SMS, Slack, Teams and social. There are no per-channel fees, so every channel runs at the same rate.
The channels page lists Voice AI, web chat, email, WhatsApp, social through Instagram and Messenger, and Slack. The pricing page adds SMS and Teams.
Voice is a real product here, with its own product page, telephony stack and SIP setup docs. The pricing page adds local numbers in 100+ countries, recording, transcription and SIP trunking. If you take support calls, that is the part of OpenCX I rate highest.
The claim across all of them is resolution. OpenCX says its Inbound agent "resolves issues end-to-end".
No channel sits behind a tier gate, voice included: "No per-seat, per-message, or per-channel fees, ever."
Since the bill is built on resolutions, adding a channel adds nothing to it.
My AskAI covers web chat, email and Slack, plus whatever your connected helpdesk routes to it, social included. It runs inside Zendesk and Intercom, Freshdesk and Freshchat, Gorgias and HubSpot. See our helpdesk integrations page.
What are the limitations of OpenCX?
TL;DR: Both deal-breakers are commercial: your rate is negotiated, and so is the definition of the thing you are charged for. After that, one knowledge index per organization and no trial on the pricing or demo pages.
Your rate is a range until you sign. OpenCX publishes $0.75 to $0.90 per resolved ticket, and the pricing page says the figure inside that band follows your monthly volume and is agreed in writing before you go live. The gap between the two ends of that band is 15 cents on every ticket the AI closes.
The billing unit is contractual too. OpenCX publishes a strict definition of a resolved ticket, then says the exact one is agreed during onboarding. On a per-outcome bill the definition of the outcome is the product you are buying, so that is the clause I read first.
Knowledge lives in one pool per account, and the documentation is blunt about it:
"One knowledge index per OpenCX org. Separate knowledge pools mean separate orgs — flag it with support if you're hitting this."
Run three brands with three sets of policies and everything the AI knows sits in one index, unless you open three accounts (we get this question on multi-brand calls). The refund guarantee has a floor too: 500 paid conversations before you can claim it.
Then there are two published helpdesk lists holding different sets. The helpdesks page names Zendesk, Intercom, Salesforce, HubSpot, Freshdesk and Twilio Flex; the documentation home names Zendesk, Front, Intercom, Gorgias and Salesforce.
Ask which list your helpdesk sits on before you assume (only Zendesk, Intercom and Salesforce appear on both).
For whoever builds the automations: the website crawler defaults to a seven-day refresh, and an AI Action gets 60 seconds before it times out.
My AskAI publishes its rate per plan on the pricing page, and you can start on it without talking to us. Multi-brand teams run a separate agent per brand, included up to 3 brands on Scale, with each extra agent priced at $49 a month.
What knowledge sources can I train OpenCX on?
TL;DR: Website crawl, Zendesk, Intercom, Front, Freshdesk, Confluence, Notion, GitBook, Shopify, a native help center, and hand-written instructions. Everything lands in one index serving both the customer-facing agent and agent assist.
The knowledge sources documentation names what OpenCX ingests: a website crawler, Zendesk, Intercom, Front, Confluence, Notion, Shopify, GitBook, Freshdesk, its own native Help Center, and AI Instructions for a specific answer or rule you want enforced. Anything else goes behind a request button that adds it to a roadmap.
You add sources at AI Training → Data Sources and check what landed at AI Training → AI Instructions. Most connectors poll every few hours. The crawler is the exception (seven days by default, 24 hours as the floor), so a same-day price change on your site reaches the AI later than a help-center edit.
The same page warns about an easy mistake: sync a Zendesk Help Center through the Zendesk connector, crawl the same help site as well, and you index every article twice. Removal behaves the way you would hope, though: taking a source out deletes its items, leaves the original system untouched, and adding it back re-pulls everything.
The same index answers customers and backs the Assist agent, so internal-only and customer-facing material live together. Check that against your own setup.
My AskAI connects Google Drive, Notion, Confluence, SharePoint, OneDrive, Dropbox, Salesforce knowledge pages and Shopify, plus your helpdesk's own help center. Where written knowledge is thin, Train on Historic Tickets drafts articles from the tickets you have already handled (the last 5,000 historic tickets by default, more on request). Private mode keeps internal material internal. See our knowledge feature page.
What features does OpenCX have?
TL;DR: Three layers: an autonomous agent gated per channel by Autopilot, an actions layer that calls your own systems, and an assist agent for your humans. Outbound has its own agent and API.
The autonomous agent
OpenCX calls its engine Agent 5. You train it once for every channel, with confidence thresholds that "route uncertain answers to a human". Autopilot is the switch deciding whether it replies for itself on a given channel, so you can run live on chat while still watching email.
Actions and workflows
This is the layer that does things: refunds, order lookups, account changes. You build actions in the dashboard as calls to your own systems.
The actions documentation adds two shortcuts for teams with engineers. You can bulk-generate actions from an OpenAPI file (the API description your developers already have). The other route is an MCP server, a standard way of exposing one tool's functions to an AI agent.
Authentication runs on org-wide variables, named secrets or per-user tokens, and each action gets 60 seconds to finish (tight, if your own API is ever slow to answer).
On top of that sits a visual workflow builder, 50+ pre-built playbooks and a public template library.
Agent assist
The Assist agent drafts replies, fills fields and summarizes threads for your human agents. OpenCX prices the whole platform under one per-resolution rate, with no add-ons or upsells listed on the pricing page.
Outbound
Proactive messaging has its own agent, product page and API endpoint, and the bill still only counts inbound tickets the AI resolves.
My AskAI does the actions layer with Tools and Tasks: Tools call your APIs, Tasks describe a multi-step procedure in plain English, and each can run on its own or propose an action for a human to approve, charged per use on top of your plan rate. On the assist side, our AI Copilot Chrome Extension carries no seat charge inside any of our integrated helpdesks, and Internal Note Replies drafts the answer onto the ticket as a private note. See our tasks and tools page.
How do I improve OpenCX responses?
TL;DR: Watch conversations live, read the impact report for where the AI gets stuck, correct it with training scenarios, and fence off topics it should never touch. Somebody owns this permanently.
OpenCX publishes a loop: monitor, find where it went wrong, fix it, check the fix.
The monitoring half is well covered. Agent 5's page names reasoning traces, confidence scores and source citations on every response, plus conversation replays, real-time metrics, error alerts and an evaluation dashboard. The docs add an inspect mode for live conversations and a debug mode for picking one apart afterwards.
The fixing half runs on two levers (both need an owner on your team). Training scenarios are the direct correction: you show the agent what the right answer looked like. Handoff topics are the fence: you name the subjects the AI must never touch, and it passes them straight to a person.
The routine in the docs is to read the impact report and work down the list of where the AI gets stuck. CSAT is collected after resolution and trended, so you can see whether last month's fixes moved anything.
Testing before customers see a change is thinner. What I want before a change goes live is a way to replay yesterday's tickets against today's configuration.
There is a test step on an individual action, and the guarantee page separates trial conversations and pre-production sandboxing from paid production. A bulk replay of past tickets against a new configuration is not published.
An unreviewed agent keeps answering from stale training while your product moves on without it. The cheapest source is the record your team already creates: the tickets you have handled.
My AskAI gives you Echo, an AI assistant in the dashboard you can ask why the agent answered the way it did and which knowledge it used. Alongside it: batch testing before you ship, guidance rules in plain English, and self-learning that drafts knowledge by comparing our reply to your agent's. See our improve pages.
What resolution rate can I expect from OpenCX?
TL;DR: OpenCX publishes 77% as its average across customers. The definition underneath it is the same one your invoice uses.
The figure OpenCX puts on its homepage, its Agent 5 page, its about page and its funding announcement is the same one:
"77% of customer queries are fully resolved by the AI without human intervention. This is the average across all customers and represents industry-leading performance."
Fully resolved without human intervention is a stricter bar than deflection, which usually just means the customer went away. It is also the bar the invoice uses.
For where the field sits, our own AI resolution rate benchmark study covers 195 rated deployments across 38 vendors. The published median sits at 70%: the mean is 66.6%, and the range runs from 15% to 98.3%.
That study is an aggregate, so it cannot settle a head-to-head. A number labeled resolution also reads differently from one labeled automation or deflection.
Every published vendor figure is a self-selected win, so take the whole set with a grain of salt. A 77% average is still strong for the category. What you get depends on your ticket mix, how good your written knowledge is, and how much of your volume is answerable without touching another system.
My AskAI publishes the whole benchmark study: 195 rated deployments, every vendor's figure, and what drives the spread. See our AI resolution rate benchmarks post.
What AI model does OpenCX use?
TL;DR: OpenCX describes a multi-model engine and names Azure OpenAI in the EU as its AI subprocessor. It does not publish which model handles which job.
The subprocessor list is the most concrete surface here, and it names one AI provider: Azure OpenAI in the EU, running in Sweden. OpenCX names a wider stack on its own site: Anthropic's Claude and OpenAI's GPT models for text, plus Deepgram and AssemblyAI for turning speech into text on the phone channel.
Agent 5 is described as "a multi-model AI with training, observability, compliance, and evaluation capabilities". Which model answers which kind of question is not published (nor is the routing logic between them).
On data handling the claims are clear. Customer data is "encrypted, isolated to your tenant, and never trains shared models", and detecting and removing personal information from conversations is a documented feature.
Get that in the contract as well as on the website.
My AskAI runs a mix of models from OpenAI, Google and Anthropic, picking each for the job it does best, with continuous A/B testing behind the scenes. Your data is never used to train our models or anyone else's.
What languages does OpenCX work in?
TL;DR: 100+ languages, with one wrinkle: phone agents carry a language setting on each agent, so a multilingual phone line needs planning.
OpenCX publishes support for 100+ languages, and its funding announcement promises customers answers in any language at any time.
Coverage varies by feature, so check before you promise anything internally. OpenCX's own documentation says the Help Center gets its own AI translation into every language you support, and your manual edits survive a re-translation. On phone agents, language is a setting on each agent, so a multilingual line needs a plan for how calls reach the right one (a Spanish caller and a German caller cannot land on the same agent).
Text channels need nothing extra. Settle voice during setup, when you decide which agent takes which language.
My AskAI supports 95 languages and detects the language on each message, so the AI replies in whatever the customer wrote in. On Intercom, our Live Translation add-on translates a handed-over conversation for your agent, at $0.05 per ticket on top of your plan.
How secure is OpenCX?
TL;DR: SOC 2 Type II and ISO 27001 certified, GDPR and EU AI Act on the footer, and an all-EU subprocessor list. HIPAA and PCI are described as ready, a different claim.
Every page footer carries the same set: SOC 2 Type II certified, SOC 2 Type I certified, ISO 27001 certified, GDPR and EU AI Act. The guarantee page repeats it and adds HIPAA and PCI.
Both are described as ready, which is a claim about capability (certified is a claim an auditor signed). If you work in a regulated industry, ask which one you are being sold.
Your security reviewer will go straight to the subprocessor list. OpenCX states on that page that "All our subprocessors are EU-based and GDPR compliant."
The list names AWS in Frankfurt, Azure OpenAI in Stockholm and Stripe in Dublin (Vercel, Cloudflare, Neon and PostHog sit in Frankfurt, Databuddy in Finland). Those are the suppliers' own compliance positions. On your vendor form they belong on a different line from OpenCX's own certifications.
On controls, OpenCX documents TLS 1.3 in transit, AES-256 at rest, automatic detection and redaction of personal and health data, full audit trails, tenant isolation, and no training of shared models on your data. Data residency lands in the EU across the board, which shortens the GDPR conversation.
My AskAI is SOC 2 Type II certified and GDPR compliant, with AES-256 encryption at rest and TLS in transit. The SOC 2 Type II report is available on Scale and Enterprise. See our security page.
Who is using OpenCX?
TL;DR: Ten named customers with their own story pages, each carrying a published automation figure, across payments, ticketing, logistics, mobility and retail.
OpenCX publishes customer stories you can open, and each one carries a number.
- Mollie, the payments platform used by 250,000+ businesses, reports 70% automation.
- TicketSwap, a ticket resale marketplace with 19M users, reports 78%.
- Trunkrs, a last-mile delivery business, reports 80% alongside a 4.8/5 customer rating.
- MoneyGram, serving 55M customers across 200+ countries, reports 80%.
I start with the industries those stories come from. Payments, marketplace ticketing and logistics have very different ticket mixes, and OpenCX publishes stories from all three.
Mollie also invested in OpenCX, which the customer story page does not mention. One of the co-founders, posting on Reddit, wrote:
"mollie is a customer (and then they invested lol)."
Customers rarely put money into a vendor they are about to churn from.
My AskAI works with 200+ ecommerce and SaaS businesses, and our agents have resolved over 1,000,000 tickets. Our case studies name the customer, the helpdesk and the numbers too: YouGarden reached 66% AI resolution on Freshdesk across about 12,000 tickets a month, TravelJoy went from 24% to 80% AI resolution on Zendesk, and RecruitCRM reached 68% on Intercom. See our case studies hub for the rest.
How much does OpenCX cost?
TL;DR: $0.75 to $0.90 per resolved ticket, nothing for tickets a human handles, no per-seat or per-channel fees, whole platform included. Your rate and the definition of a resolution are both agreed in writing.
I like that the rate sits in the title of OpenCX's own pricing page. The published range lets you bracket a cost before you book a demo.
What OpenCX charges for
One rate covers the outcomes the AI delivers, at $0.75 to $0.90 per resolved ticket, set by your monthly volume, with higher volume pulling the rate down. Above 100k resolutions a month the published range stops applying and you get a custom quote. Tickets your team handles are free:
"If Open hands a ticket to one of your agents, you're not billed for it."
There are no setup fees, no monthly platform fee, no per-seat charge and no per-message charge. The one thing billed outside the rate is carrier minutes on the phone channel, passed through at cost and typically under $0.10 per resolution.
The rate buys everything else: every channel, an AI-native helpdesk with unlimited seats if you want it, the telephony stack, outbound, workflows, knowledge management and implementation support (the pricing page's own phrase is "no add-ons, no upsells").
What counts as a billable outcome
OpenCX's published definition:
"A ticket Open closes end-to-end without a human teammate stepping in, where the customer either confirms the outcome or doesn't re-open it. The exact definition is agreed in writing during onboarding."
The first half is a strict, defensible definition. The closing sentence makes it negotiable, so what counts on your account is whatever your onboarding document says.
Get that onboarding document, read the edge cases, and ask what happens to a ticket the customer reopens a week later.
The $2M guarantee
OpenCX backs deployments with a refund promise. Within the first 120 days you can claim back everything you spent, on satisfaction alone with no metric to hit, capped at $2M and paid within 30 days (you need 500 paid conversations on the clock first). It is the closest thing OpenCX offers to a trial.
What it costs at 2,000 conversations a month
Assumptions, applied identically to both rows:
- 2,000 customer conversations a month
- 77% handled by the AI, which is OpenCX's published average, so 1,540 resolved
- No per-seat charges on either side
- No voice carrier minutes included
Vendor | Plan priced | Monthly cost | Notes |
OpenCX | Single per-resolution rate | $1,155 to $1,386 | 1,540 resolutions at $0.75 to $0.90. Tickets your team handles are not billed. Carrier minutes extra on voice |
My AskAI | Pro | $319 | $199 base including 1,000 credits, then 1,000 more at $0.12. Billed per ticket, resolved or not. Pro covers 5 team seats |
OpenCX's bill rises with every extra ticket the AI gets right: lift the AI's share from 1,540 of those 2,000 conversations to 1,800 and the same month costs $1,350 to $1,620. We charge for the tickets that arrive, so ours stays at $319 whatever the resolution rate does.
My view on outcome pricing comes down to attribution. Most of what lifts a resolution rate is work you do:
- knowledge updates
- tool and API connections
- guidance tuning
- triage setup
- the weekly QA loop
The vendor brings model choice and architecture. Outcome pricing charges you more as your own work pays off, which is why we price on the unit you already track and run a 30-day trial with everything unlocked, no card and no volume limit. We have written separately on outcome-based pricing and on per-conversation versus per-resolution billing.
My AskAI is $199 a month on Pro with 1,000 credits included and $0.12 a credit after that, or $499 on Scale with 2,000 credits and $0.10 after. You pay per ticket, resolved or not, and the rate does not climb as the AI gets better. See our pricing page, our pricing explainer and the ROI calculator.
Does OpenCX have a free trial?
TL;DR: No trial on the pricing or demo pages. You can open an account for the web widget, but the buying route is a demo and a quote, with the 120-day refund standing in.
The pricing and demo pages publish no trial. There is no trial row on the pricing page, and the buttons all point the same way: book a demo, get started, contact sales. The demo page is a request form.
You can still open an account yourself. The Quickstart's first step is to do exactly that, and you can have a web widget answering questions in minutes. What that account includes before somebody calls you is not published (treat it as a first look).
The evaluation route runs: demo, quote, signed rate, deploy, and 120 days in which you can ask for your money back. That is a real answer to whether you can try it, and you pay first either way. Finance signs off the spend before you have any evidence from your own tickets.
My AskAI gives you 30 days free with every feature unlocked, no ticket limit and no credit card. You sign up with a work email and see a real month of your own volume before you pay us anything.
Is OpenCX worth it?
TL;DR: Yes for a high-volume team running voice and text that will sign a commercial agreement. No for one that needs a published rate and a self-serve pilot to build the case.
OpenCX is a strong product, and its commercial terms decide whether it fits your team.
If you land in the second list, our roundup of AI help desk software covers the wider field.
This verdict rests on fourteen things: what OpenCX is, setup, channels, limitations, training sources, features, improvement, resolution rate, model choice, languages, security, customers, cost and the trial. Evidence comes from OpenCX's own product and pricing pages, its documentation, its customer story pages, its subprocessor list, and our benchmark study of published resolution rates. Where OpenCX publishes no figure, the post says so.
Choose OpenCX if:
- You handle real volume across more than one channel and want voice and text answered by the same trained agent
- You already run Zendesk, Intercom, Salesforce, HubSpot or Freshdesk and will not migrate
- You can sign a commercial agreement and negotiate both a rate and a resolution definition
- You want the vendor carrying deployment risk, which the 120-day refund does properly
Don't choose OpenCX if:
- You need to run a real pilot on your own tickets before anyone signs
- You need a published per-tier rate to build a business case without a sales cycle
- You run several brands and need separated knowledge pools inside one account
- Your support is text-only and low volume, where a per-resolution rate has less to work with
My AskAI is built for the second list: a published rate, a 30-day trial on your own tickets, a separate agent per brand, and our install running inside the helpdesk you already have.
What are the Pros and Cons of OpenCX?
Pros
- Billed on the outcomes the AI delivers: $0.75 to $0.90 per resolved ticket, nothing charged when a ticket reaches one of your agents, and no per-seat, per-message or per-channel fees.
- One agent covers every channel, and the platform comes with it: voice, chat, email, WhatsApp, SMS, Slack, Teams and social run off the same training, with the helpdesk, telephony, outbound and workflows at the same rate.
- The customer proof is checkable at depth: ten story pages you can open, each naming the customer and its automation figure, across payments, ticketing, logistics, mobility and retail.
Cons
- The rate is a published range: $0.75 to $0.90 is a 20% spread agreed in writing before you go live, so your business case stays provisional until sales confirms it.
- The billing unit is negotiated too: OpenCX's definition of a resolved ticket ends with that definition being agreed during onboarding, so the thing you pay for is a contract term.
- No trial on the pricing or demo pages: evaluation runs from demo to paid deployment, and the 120-day refund only becomes available after 500 paid conversations.
- One knowledge index per organization: multi-brand teams who need separated knowledge pools have to run separate accounts.
OpenCX
- Brand: OpenChat, Inc.
- Rating: 7.5/10
- In a sentence: Aligned pricing, checkable customer proof, and one agent covering voice and text with the whole platform included. It is held back from an 8 or 9 by a rate and a resolution definition you cannot know until you have negotiated both.
The commercial model earns most of that number from me. The rate is published as a range, nothing is charged when a human takes over, and a $2M refund sits behind the deployment.
The customer proof is ten named story pages, each carrying its own automation figure. It is one system at one rate, with reasoning traces and conversation replays you can open.
Two things stay invisible until you are in a sales conversation: the price a given buyer pays, and the definition of the thing being priced.
We publish both: the rate per plan, and what counts as a billable ticket. You can also run 30 days on your own tickets before you commit.
FAQs
What is OpenCX?
OpenCX is an AI customer support platform made by OpenChat, Inc. A single trained agent handles voice, chat, email, WhatsApp, SMS, Slack and social on top of the helpdesk you already use, billed per ticket the AI resolves end to end, with no per-seat fee.
Which company is behind OpenCX?
OpenChat, Inc. makes it, and the open.cx platform is where its pricing, documentation and customer stories live. Y Combinator backs the company, and a $7M round has been announced led by YC and X By Unifonic (Shorooq and Sadu also took part).
Does OpenCX replace my helpdesk or sit on top of it?
Either, though the usual deployment is on top: OpenCX's line is that it sits on top of the stack you already run, with no migration. Its own AI-native helpdesk is included in the rate for teams that do want to move.
Which helpdesks does OpenCX integrate with?
OpenCX publishes two lists. Its helpdesks page names Zendesk, Intercom, Salesforce, HubSpot, Freshdesk and Twilio Flex. Its documentation home names Zendesk, Front, Intercom, Gorgias and Salesforce. Check that yours is on the list matching what you are buying (the two overlap on Zendesk, Intercom and Salesforce only).
What does OpenCX cost and how does it count an outcome?
$0.75 to $0.90 per resolved ticket, set by monthly volume, with nothing charged for tickets your team handles and no per-seat, per-message or per-channel fees. A resolution is a ticket the AI closes end to end without a human stepping in, and the exact definition is agreed in writing during onboarding (above 100k resolutions a month the rate is quoted individually).
Does OpenCX do voice as well as chat and email?
Yes. Voice is a full channel with its own product page, telephony stack and documentation, covering local numbers in 100+ countries plus recording, transcription and SIP trunking. Carrier minutes are passed through at cost.
What resolution rate does OpenCX publish?
77%, described as the average across all its customers and defined as queries fully resolved by the AI without human intervention. Our own benchmark study of 195 rated deployments across 38 vendors puts the published median at 70%, with the range running from 15% to 98.3%.
What knowledge sources can OpenCX be trained on?
A website crawler, Zendesk, Intercom, Front, Freshdesk, Confluence, Notion, GitBook, Shopify, its own native Help Center, and hand-written AI Instructions. Everything lands in a single index per organization, serving both the customer-facing agent and the assist agent.
What if we have no help center to train an AI agent on?
Knowledge that never became a help-center article usually sits in tickets instead. With My AskAI, our Train on Historic Tickets feature reads the conversations your team has already handled and drafts knowledge articles from them for review (the last 5,000 historic tickets by default, more on request).
What AI model does OpenCX use?
OpenCX describes Agent 5 as a multi-model AI. Its subprocessor list names Azure OpenAI in the EU as its AI provider, and its own site adds Anthropic's Claude and OpenAI's GPT models plus Deepgram and AssemblyAI for speech. Which model handles which task is not published.
Does OpenCX have a free trial?
No trial on the pricing or demo pages. OpenCX offers a satisfaction-only refund of everything you have spent within the first 120 days, capped at $2M, once you have run at least 500 paid conversations. You can also create a self-serve account and get a web widget answering questions quickly (what that account includes is not published).
Is OpenCX worth it for a mid-market support team?
If you run real volume across voice and text and can sign a commercial agreement, yes: the pricing is aligned, the platform is included in the rate, and the customer proof is unusually checkable. If you need a published rate and a self-serve pilot to make the case internally, the buying process will get in your way first.
