13 Best Chatbot APIs for Websites in 2026 (Free and Paid Compared)

Key takeaways

Search for a chatbot API and you find two different products. Model APIs return text and leave the widget, the inbox and the human handoff to you. Chatbot platforms ship all of that and offer an API on top, often only on a higher plan. Pick the wrong kind and you rebuild a support stack from scratch or pay for an API that can’t do what you need. We compared 13 chatbot APIs on their docs and pricing pages, noting which plans include API access, what each bills for and who should skip it.

Tidio is listed first because the list is ordered for teams adding a chatbot to a website or store. Every tool gets the same criteria and its real limits, Tidio included.

The best chatbot APIs at a glance

Our top three picks: Tidio for small teams and online stores that want a finished chatbot, OpenAI for teams building a custom bot on a model API, and Rasa for teams that must self-host.

Tool	Best for	Free option	Starting price
Tidio	Small teams and online stores	Free plan, 7-day trial	$24.17/mo billed annually ($29 monthly)
OpenAI	Custom bots on GPT models	No free credit (prepaid, $5 minimum)	$0.10 / $0.50 per 1M input / output tokens (GPT-6 Luna)
Anthropic Claude	Long-context and careful-reply bots	No free tier (prepaid credits)	Per token, by model
Google Gemini	Multimodal bots and prototypes	Free tier on selected models	$0.25 / $1.50 per 1M input / output tokens (3.1 Flash-Lite)
Dialogflow CX	Voice and multi-channel flows	$600 trial credit for new customers	$0.007 per text request
Botpress	Developer-built agents	Free plan (100 conversations)	$150/mo billed annually
Rasa	Self-hosted, code-first teams	Open-source core and a free Developer Edition	Sales-led (Rasa Pro)
IBM watsonx Assistant	Regulated industries	Lite plan, 10,000 message calls a month	Custom (Plus and Enterprise)
Intercom Fin	Support teams on Intercom	No	$0.99 per resolution, plus seats
HubSpot	Teams already on HubSpot	Free tools for up to 2 users	$7/seat/mo billed annually ($20 monthly)
Wati	WhatsApp-first stores and support	7-day trial	$59/mo billed annually ($69 monthly)
ManyChat	Instagram and Messenger automation	Free plan (25 active contacts)	$14/mo
Ada	Large multilingual support teams	No	Sales-led, no public list

Prices change often. Confirm each one with the vendor before you commit.

How we chose these chatbot APIs

We started from the APIs that Google roundups, developer threads and AI answers name most for website chatbots, and kept 13 that are still sold and still do what we describe. We judged each on the same criteria:

• Fit. The order follows fit for teams adding a chatbot to a website, not G2 score.
• What the API does. Model text, or contacts, conversations and tickets.
• What is built in. Widget, inbox, handoff and analytics, or none.
• Plan gating. Which plan unlocks the API, and its rate limits.
• Billing unit. Tokens, requests, seats, contacts or resolutions.
• Free entry. A free plan, a trial or neither.

What is a chatbot API?

A chatbot API is an interface that lets your site or app send a message to a conversational engine over HTTPS and get a structured reply back. The engine is either a language model or a rule-based bot, and the reply can be text, a suggested action or a call to one of your own tools.

Every chatbot API follows the same five steps:

• Authenticate. With an API key, a client ID and secret pair, or an OAuth token.
• Request. Send the user’s message, the history and usually a system prompt.
• Process. The provider runs it through its model or rule engine.
• Respond. You get text, actions or tool calls back, often streamed.
• Notify (optional). Webhooks push events, such as a handoff request, to your URL.

What the first call looks like

A model API takes a prompt and returns text. This Python call to OpenAI’s Responses API is the smallest working example, and the Anthropic and Gemini SDKs follow the same pattern:

from openai import OpenAI

client = OpenAI()  # reads OPENAI_API_KEY from the environment
reply = client.responses.create(
    model="<model-name>",
    input="What is your return policy?",
)
print(reply.output_text)

A platform API usually manages the data around the bot instead. This curl call lists the operators in a Tidio project with the client ID and secret you generate under Developer > OpenAPI:

curl -H "X-Tidio-Openapi-Client-Id: <YOUR_CLIENT_ID>" \
     -H "X-Tidio-Openapi-Client-Secret: <YOUR_CLIENT_SECRET>" \
     -H "Accept: application/json; version=1" \
     https://api.tidio.com/operators

Build or buy a chatbot API?

Two paths lead to a working chatbot. The right one depends on how much custom logic you need and how much engineering time you have.

Scenario	Buy a platform	Build on a model API
You need a working widget this week	Yes	No
You have little engineering time	Yes	No
You need custom retrieval or multi-step agents	Some platforms	Yes
You need strict data residency	Depends	Yes, with self-hosting

If you can describe the bot’s job in a few sentences, buy a platform. If it has to reason over your own data or call your own tools, build on a model API or a framework. Our chatbot framework guide covers the build side.

The 13 best chatbot APIs in 2026

1. Tidio

Best for: Small teams and online stores that want a finished chatbot with an API around it.

Tidio is the managed route. The live chat widget, Flows and Lyro, the AI agent that can resolve up to 75% of support requests, work out of the box, so the API extends the bot instead of powering it.

Plan gating is the catch: OpenAPI isn’t available on Free, and webhooks are Plus and Premium only. The APIs also manage data around the bot. The OpenAPI has a Lyro endpoint for answering tickets, but it isn’t a general model endpoint, so if you want to own every prompt, pick a model API.

Key features:

• Widget SDK: frontend scripts that identify visitors, set contact properties and add custom triggers.
• OpenAPI: reads and writes live data for CRM sync, tickets and store or payment integrations.
• Webhooks: events pushed to your endpoint, with signature verification.
• Analytics API: read-only historical data for Looker, Power BI or a warehouse, on the same credentials.

Tidio fits these teams best:

• Online stores, especially Shopify stores, that want live cart preview and order history in the chat.
• Teams that handle live chat, email, Messenger, Instagram and WhatsApp in one inbox.

Pros

• Widget, AI agent and inbox are included, so you can launch before you write code.
• Four developer surfaces cover frontend, data, reporting and events.
• Native Shopify and WordPress integrations alongside the API.

Cons

• OpenAPI and webhooks start on the Plus plan, so smaller teams use the Widget SDK and built-in integrations first.
• Plans below Plus can call the product endpoints only, at up to 10 requests a minute.
• Lyro is priced separately, so you only pay for what you use, but there are conversation quotas to keep an eye on.

Skip Tidio if you want to run your own model, prompts and hosting.

Pricing: Free plan with 50 billable conversations a month and up to 10 seats, no OpenAPI. Starter $29 a month ($24.17 billed annually). Growth from $59 ($49.17 annually). Plus from $300 plus usage, with OpenAPI at 60 requests a minute and webhooks. Premium through sales, at 120 requests a minute. Lyro AI Agent is sold separately from $39 a month ($32.50 annually) for 50 conversations, and every account gets 50 Lyro conversations once. 7-day free trial. See Tidio pricing.

Rating: G2 4.6/5 (1,995 reviews), G2 reviews.

Read more: Tidio developer docs.

2. OpenAI

Best for: Teams building a custom chatbot on GPT models.

OpenAI recommends its Responses API for new chatbot and agent builds, with conversation state, retrieval through file search and tool calls. It has the biggest tutorial and tooling ecosystem, and everything above the model is yours to build.

Key features:

• Responses API: one request shape for text, tools and state.
• Tool calling: let the bot look up orders through your own functions.
• SDKs: Python, JavaScript, .NET, Go, Java and Ruby.

Pros

• Largest ecosystem of examples and libraries.
• Several model sizes to match cost to the task.

Cons

• No widget, inbox or human handoff.
• Models and endpoints change often, so pin your SDK version.

Pricing: billed per token on prepaid credit, with a $5 minimum first purchase and no standard free credit. GPT-6 Astra costs $10 input and $50 output per 1M tokens, and GPT-6 Luna costs $0.10 and $0.50. The Batch API takes 50% off asynchronous requests. See OpenAI API pricing {nofollow}.

3. Anthropic Claude

Best for: Bots that read long documents and need steady replies.

Anthropic’s Messages API offers system prompts, tool use, streaming and context windows of up to 1 million tokens, depending on the model. Like OpenAI, it’s a model API: you bring the widget, inbox and handoff.

Key features:

• Messages API: send a conversation and a system prompt, get a reply.
• Tool use: let the model call your functions.
• Prompt caching: repeated context can cost up to 90% less on most models.

Pros

• Handles long help-center articles and policy documents.
• Caching can cut the cost of a repeated knowledge base.

Cons

• No widget, inbox or handoff.
• Usage-based billing makes monthly cost harder to predict.

Pricing: billed per token, by model, on prepaid credits, with no general free tier. Prices differ by model, so read the current table before you size a budget. See Claude API pricing {nofollow}.

4. Google Gemini API

Best for: Prototypes and multimodal bots that need a free starting point.

The Gemini API offers Google’s models through AI Studio and Vertex AI, with a free tier on selected models. Limits vary by model and project and are shown in AI Studio. It suits prototypes and bots that read images or audio as well as text.

Key features:

• Free tier: test the models in AI Studio.
• Multimodal input: text, images and audio in one request.
• Function calling: let the model call your tools.

Pros

• A free tier makes testing cheap.
• Production path through Google Cloud.

Cons

• Free-tier terms differ from paid use, so read them before you send real customer data.
• No widget, inbox or handoff.

Pricing: free tier with rate limits, then pay per token. Gemini 3.1 Flash-Lite costs $0.25 input and $1.50 output per 1M tokens. See Gemini API pricing {nofollow}.

5. Google Dialogflow CX

Best for: Voice and multi-channel flows on Google Cloud.

Dialogflow CX, now called Conversational Agents, is Google’s visual builder for complex flows. It supports voice and phone natively, and its generators use models from Vertex AI. Its state-machine model suits structured tasks and takes more setup than a drop-in widget.

Key features:

• Visual flow builder: design conversations as pages and routes.
• Voice and IVR: native phone support.
• Vertex AI link: add generative answers to scripted flows.

Pros

• Good fit for voice and multi-step flows.
• Sits inside Google Cloud security and IAM.

Cons

• Two products (ES and CX) confuse newcomers.
• Per-request pricing needs capacity planning.

Pricing: billed per request, not per session. Text requests cost $0.007, generative AI requests $0.012 and voice $0.06 a minute. New customers get a $600 trial credit, valid for 12 months. See Dialogflow pricing {nofollow}.

6. Botpress

Best for: Developer-built agents with a hosted runtime.

Botpress is an agent platform for developers: you build the bot in its studio, call it over an HTTP Chat API and connect channels through integrations. You can extend it with custom JavaScript or TypeScript. It sits between a model API and a finished platform.

Key features:

• Studio: build workflows and knowledge bases visually.
• Chat API: call the bot from your own app.
• Integrations: WhatsApp, Messenger, Slack, Teams, Shopify and Zapier.

Pros

• Custom code inside the bot.
• Hosted runtime, so you don’t build the chat layer.

Cons

• Usage-based plans mean the bill moves with traffic.
• Less built-in inbox and handoff than a support platform.

Pricing: free plan with 100 conversations, 3 seats and 3 AI agents. Plus is $150 a month billed annually with 250 conversations, and extra usage is charged per conversation. We don’t link Botpress’s pricing page; check it with the vendor.

7. Rasa

Best for: Code-first teams that need self-hosting and full data control.

Rasa is a well-known open-source conversational AI framework. The core is free under the Apache 2.0 license, and the enterprise platform adds analytics and a commercial license. It suits Python teams with a strict data-residency story and needs engineering time to run.

Key features:

• Self-hosting: run it in your own cloud or on-prem.
• Python-first: customize every layer in code.
• Bring your own LLM: mix generative answers with scripted flows.

Pros

• Full data ownership and deployment control.
• Free open-source core and Developer Edition.

Cons

• Needs Python engineers to build and maintain.
• No hosted widget or inbox.

Pricing: the open-source core is free to self-host. The enterprise platform is sales-led, priced by annual conversation volume and deployment, and you also pay for hosting, model usage and engineering time. See Rasa {nofollow}.

8. IBM watsonx Assistant

Best for: Regulated industries that need on-prem or hybrid deployment.

watsonx Assistant offers enterprise governance and a choice of managed cloud or on-prem deployment, including IBM Cloud and AWS. IBM advertises a HIPAA-ready option, with coverage depending on plan and deployment. IBM’s pricing page now sits under watsonx Orchestrate, so confirm current plan names before you buy.

Key features:

• Deployment choice: managed cloud or on-prem.
• Governance: audit and access controls.
• SDKs: Java, Node.js, Python, Ruby, .NET, Go, Swift and Unity.

Pros

• A free Lite plan for prototypes.
• On-prem and hybrid options.

Cons

• Steeper learning curve than SMB tools.
• IBM no longer publishes a Plus price, so budgeting means a call with sales.

Pricing: Lite is free with 10,000 API message calls a month (about 1,000 monthly active users), though IBM now limits new Lite instances. Plus and Enterprise are billed by monthly active users. See watsonx Assistant pricing {nofollow}.

9. Intercom Fin

Best for: Support teams that already run on Intercom.

Fin is Intercom’s AI agent, layered on its support platform. You configure Fin inside Intercom, and the REST API covers contacts, conversations and tickets. If your team already lives in Intercom, Fin plugs into your macros and reporting. Costs climb with volume, because Fin bills per resolution on top of seats.

Key features:

• Fin AI agent: answers from your help content and hands over to agents.
• REST API: contacts, conversations and tickets.
• Reporting: resolution and deflection rates.

Pros

• Mature support tooling: SLAs, macros and workflows.
• Detailed reporting on AI resolutions.

Cons

• Per-resolution fees add to seat prices.
• Heavier than most small teams need.

Pricing: Fin costs $0.99 per resolution on top of seats. Intercom’s own pages show different starting seat prices ($19 and $29 a month on annual billing), so confirm the figure for your account. We don’t link Intercom’s pricing page; check it with the vendor.

10. HubSpot

Best for: Teams already invested in the HubSpot CRM.

HubSpot’s Conversations API lets you send and receive messages, manage threads and trigger workflows from chat events. The free tools include limited chatflows, and some AI features are free while others need paid plans. Its value compounds when contacts, deals and chat already live in HubSpot.

Key features:

• Conversations API: messages, threads and workflow triggers.
• Chatflows: rule-based bots in the chat tool.
• CRM sync: contacts and deals update from chat.

Pros

• Free tools with API access.
• Two-way CRM sync.

Cons

• Bot logic is rules-based by default; AI answers need paid plans or a connected model.
• Best value only if you already use HubSpot.

Pricing: Service Hub is free for up to 2 users. Starter is $20 per seat a month billed monthly ($7 on annual billing). Professional is $100 per seat monthly ($90 annual) plus a one-time $1,500 onboarding fee. We don’t link HubSpot’s pricing page; check it with the vendor.

11. Wati

Best for: WhatsApp-first stores and support teams.

Wati wraps the WhatsApp Business API in a friendlier developer surface. The REST API covers messaging, templates, contacts and webhooks. It’s a good fit when your customers are already on WhatsApp, and a poor one when you need a web widget.

Key features:

• Templates and opt-ins: send approved templates and manage consent.
• REST API: messaging, contacts and webhooks.
• Broadcasts: message contact segments in bulk.

Pros

• Built around WhatsApp Business.
• Quick setup for high-volume messaging.

Cons

• WhatsApp only, with no web widget or unified inbox.
• WhatsApp message fees are charged on top of the plan.

Pricing: 7-day free trial. On annual billing, Growth is $59 a month, Pro $119 and Business $279. WhatsApp message fees are charged separately. We don’t link Wati’s pricing page; check it with the vendor.

12. ManyChat

Best for: Instagram, Messenger and WhatsApp marketing automation.

ManyChat focuses on social-first flows: opt-ins, broadcasts, comment automations and quick replies. Marketers can ship campaigns without engineers. It’s less suited to support workflows or a deep developer surface.

Key features:

• Comment automations: trigger a DM from an Instagram comment.
• Broadcasts: message opted-in contacts.
• Quick replies: guide users with buttons.

Pros

• Deep Instagram and Messenger support.
• A free plan to test the platform.

Cons

• The free plan is a small sandbox.
• Limited for complex support workflows.

Pricing: the free plan covers 25 active contacts, one user and up to two channels, with ManyChat branding. Essential is from $14 a month for 250 contacts, Pro from $29 for 2,500 and Business from $69 for 7,500. ManyChat’s pricing and AI pages disagree on whether AI is included in Pro, so confirm it for your account. We don’t link ManyChat’s pricing page; check it with the vendor.

13. Ada

Best for: Enterprise self-service automation across many languages.

Ada is a no-code AI agent platform with an API for conversations and analytics. Generative and scripted flows sit behind the guardrails that enterprises expect. It’s built for large teams, so smaller ones will find it heavy and slow to deploy.

Key features:

• Generative and scripted flows: with policy guardrails.
• Multilingual support: many languages from one agent.
• Enterprise controls: SSO and audit logs.

Pros

• Guardrails and controls for enterprise teams.
• Strong multilingual support.

Cons

• Pricing is sales-led, with no public list.
• Longer time to value than SMB-focused tools.

Pricing: sales-led. Ada doesn’t publish rates and prices mainly by conversation, and a Salesforce AgentExchange listing shows $30,000 a year as a starting price. We don’t link Ada’s pricing page; check it with the vendor.

Free chatbot APIs worth trying

“Free” means different things here: a free plan with caps, a free model tier or an open-source core.

Tool	What you get free	What is paid-only	Good for
Tidio	Live chat and 50 Lyro conversations once, no OpenAPI	OpenAPI and webhooks (Plus), more Lyro conversations	Small sites that want a working bot first
Google Gemini	Free tokens on selected models, with rate limits	Higher limits	Prototypes and multimodal bots
Botpress	100 conversations, 3 seats, 3 AI agents	Plus plan, extra conversations	Developers testing an agent platform
HubSpot	Free tools for 2 users, with chatflows and the Conversations API	AI answers, advanced automation	Teams already on HubSpot
watsonx Assistant	10,000 message calls a month on Lite, new instances limited	Plus and Enterprise	Testing enterprise NLU
ManyChat	25 active contacts, 1 user, 2 channels	Paid plans, AI features	Testing social flows
Rasa	Open-source core and a free Developer Edition	The enterprise platform	Python teams that self-host

For a free route with no engineering, start with Tidio’s free plan or HubSpot. For raw model experiments, start with Gemini. Free tiers are for testing, so check the terms before you send real customer data through one.

How to choose a chatbot API

Start with who builds the bot. A support lead or marketer should pick a platform such as Tidio or HubSpot, and an engineering team has room to use a model API or a framework such as Rasa. Then match four things:

• Channels. A web-only bot fits Tidio, Intercom or HubSpot. WhatsApp points to Wati, social DMs to ManyChat, and voice to Dialogflow CX.
• Workflow. Standard support fits a platform. Custom lookups, transactions or reasoning over your own data need a model API or a framework.
• Security. Check HIPAA, GDPR and SOC 2 coverage with each vendor, and prefer self-hosted Rasa for strict residency. Model APIs add their own risks, so plan for prompt injection prevention as well as compliance.
• Cost shape. A subscription gives a predictable bill, and per-token billing scales with traffic, so set a cap in week one. Our guide to what chatbots cost goes deeper.

Before you decide, run a one-week pilot on your own top 20 customer questions.

Which chatbot API should you pick?

• You run an online store and want a chatbot this week: Tidio’s Growth plan with Lyro, then Plus if you need OpenAPI and webhooks. On Shopify you also get live cart preview and order history in the chat.
• You have developers and want control of the model: OpenAI, Anthropic Claude or Google Gemini, with your own widget and handoff.
• You need voice or phone flows: Dialogflow CX.
• You must self-host or keep data on your servers: Rasa, or watsonx Assistant in regulated industries.
• Your customers live on WhatsApp or Instagram: Wati or ManyChat.
• You already run support on Intercom or sales on HubSpot, or you are a large multilingual team: Intercom Fin, HubSpot or Ada.

Frequently asked questions