Custom AI chatbot integration for Salesforce
We build custom AI chatbots that work with Salesforce Service Cloud. They read and update cases, hand work to your reps through Omni-Channel, and can call an Agentforce agent when that is the better tool.
Updated
Cases your reps can work
The chatbot creates or updates the case, and Omni-Channel routes it to the right rep.
Agentforce where it fits
Your chatbot can hand Salesforce tasks to an Agentforce agent through the Agent API.
Engineers from $45 an hour
Published rates, and a written quote after discovery for a defined project.
How does a custom chatbot connect to Salesforce?
In one of three ways. It can read and write Service Cloud records such as cases through Salesforce's REST API. It can call an Agentforce agent through the Agent API. And inside Salesforce, Flow and Apex can call your own model. We usually start with the REST API and add the others where they help.
Each route authorizes through an external client app or a connected app with an OAuth flow, and Salesforce's REST API guide says creating new connected apps is restricted as of Spring '26, recommending external client apps instead (Salesforce REST API guide). How any project runs is on our AI chatbot integration page; this page covers Salesforce.
How does the Salesforce Agent API work?
Your application gets a token from an external client app, starts a session with an active Agentforce agent, sends messages synchronously or as a stream, and closes the session when the conversation is over. Each call has a 120-second timeout, usage draws on your Agentforce credits, and agents of the type Agentforce (Default) aren't supported.
Salesforce's Trailhead sets out the lifecycle (Trailhead):
- Enable Agentforce and activate at least one agent.
- Create an external client app with OAuth settings. It handles the client credentials flow and holds the consumer key and secret.
- Request a token with the consumer key, consumer secret and your domain name. Every Agent API call needs one.
- Start a session with your own session key; the response returns a session ID and the agent's greeting. Send messages synchronously or streamed for live chat, then close the session.
The client credentials flow has no person signing in, so calls run as an execution user that an admin enters in the external client app's policies (Trailhead), often called the Run As user. That user's permissions decide what the integration can see and change, so we use a dedicated user with only the access it needs.
The Agent API puts an Agentforce agent behind a channel or app you build; Salesforce's example is your website. It doesn't run your own model. We use it when a custom chatbot should hand a Salesforce task to an Agentforce agent.
Our rates
Published prices in US dollars. Monthly plans are 160 hours, about 15% below the hourly rate.
Mid-level developer
3 to 5 years
$25 / hour
$3,400 a month (160 h)
Senior developer
6 or more years
$35 / hour
$4,750 a month (160 h)
AI / LLM engineer
Agents, RAG, IBM watsonx
$45 / hour
$6,100 a month (160 h)
Senior AI engineer
Leads AI architecture and delivery
$60 / hour
$8,150 a month (160 h)
Two-week trial with the developer you choose. No charge if you stop within the first five working days. US Eastern clients get 09:00 to 12:00 US Eastern time through an evening shift, at 15% extra. Full pricing and comparisons.
Can Salesforce Flow and Apex call an agent or your own model?
Yes, both. Flow can call an Agentforce agent as an action: Salesforce's Metadata API lists generateAiAgentResponse among Flow action types. Apex can run invocable actions through its Invocable.Action class. To call your own model instead, External Services turns your model's API into Flow actions, and Apex can make HTTP callouts.
- Agent from Flow. The Metadata API guide describes generateAiAgentResponse as an action that "generates a response from the AI agent based on input and instructions", from API version 63.0 (Salesforce).
- Actions from Apex. The Apex Reference Guide's Invocable.Action class creates and runs invocable actions from Apex code, including custom actions by type and name (Salesforce).
- Your model from Flow. External Services registers an outside REST API from its OpenAPI specification and turns its operations into invocable actions for Flow Builder, without code (Trailhead).
- Your model from Apex. Apex sends HTTP requests to an outside service once its endpoint is authorized (Trailhead).
Salesforce's InvocableMethod annotation makes Apex callable "from declarative tools like Flows, external applications via REST, and Agentforce" (Trailhead), so a rule written once can serve a custom chatbot and Agentforce alike.
How does a custom chatbot work with Service Cloud cases?
Salesforce's object reference defines a case as a customer issue or problem. Through the REST API the chatbot can look up the customer's open cases, add what it learned, or create a new case with a summary. Omni-Channel then routes the case to a rep by availability, skills and workload, so your existing queues stay in charge.
The Case object supports create, update, upsert, query and search calls, among others (Salesforce Object Reference), and the REST API creates and updates records through its sObject resources (Salesforce). The chatbot works as its integration user, so case access follows that user's permissions. A typical flow:
- Identify the customer and find their contact record and any open case.
- Answer what it can, and record the outcome on the case.
- When a person is needed, update or create the case with a summary and what the chatbot already tried, and let Omni-Channel route it.
Omni-Channel routes work items such as cases, chats and messaging sessions to the most appropriate rep in real time, by availability, skill set and workload. Salesforce is retiring Standard Omni-Channel in Summer '26, so new routing is built on Enhanced Omni-Channel (Trailhead).
How is Agentforce billed?
Three ways. Flex Credits cost $500 per 100,000, and each action uses 20 credits, which Salesforce's own examples price at $0.10. Conversations cost $2 each and are aimed at customer-facing agents. Agentforce add-ons for Sales, Service and Field Service cost $125 per user a month for employee use. Agent API usage draws on credits.
| Option | Price | What it covers |
|---|---|---|
| Flex Credits | $500 per 100,000 credits | Actions, such as updating a record: 20 credits each, 30 for voice |
| Conversations | $2 per conversation | Flat pricing, optimized for external, customer-facing agents |
| Agentforce add-ons | $125 per user a month | Unmetered Agentforce use for employees |
| Salesforce Foundations | $0 | A free way to start with Agentforce |
From Salesforce's Agentforce pricing page, checked 5 November 2026. Salesforce says the page is for information only and subject to change.
Unused Flex Credits don't roll over into the next term, and usage above your entitlement is billed monthly in arrears at your contracted rate (Salesforce). By our arithmetic, a conversation in which the agent takes three actions uses 60 credits, or $0.30 at list price.
When is Agentforce the better buy?
When your service team works in Salesforce, conversations run on Salesforce's own messaging channels, and the answers and actions live in Salesforce data, flows and Apex. Agentforce routes escalations through Omni-Channel with the conversation history attached. A custom chatbot is worth it when Salesforce is one system among several.
In Salesforce's messaging setup, one Omni-Channel flow routes each new conversation to the AI agent and another routes escalations to a queue of reps with the conversation history. If no rep is available, the agent resumes, and escalation is tried once per session (Trailhead).
- Choose Agentforce when your channels are Salesforce messaging channels, your reps work in Service Cloud, and the actions you need already exist as flows or Apex.
- Choose a custom chatbot when customers reach you in apps or channels outside Salesforce, when it must act in other systems as well as on cases, or when the model must run in your country or on your servers (private AI).
- Or combine them: the custom chatbot owns the conversation and calls an Agentforce agent for Salesforce tasks.
If Agentforce alone covers your need, we will say so on the first call.
What does a Salesforce chatbot integration cost?
Two parts: Salesforce's own costs, such as Agentforce credits if the chatbot calls an Agentforce agent, and the build. By our planning estimates, a chatbot that reads your business systems takes 480 to 960 hours, and one that takes actions such as updating cases takes 800 to 1,600. Engineering is $45 to $60 an hour.
| What the chatbot does | Build hours | At Webify.AI rates |
|---|---|---|
| Reads Salesforce and your business systems | 480 to 960 | $22,800 to $45,600 |
| Also takes actions, such as creating or updating cases | 800 to 1,600 | $42,000 to $84,000 |
Planning estimates, not quotes. Maintenance is 20 to 40 hours a month in the first months, $1,050 to $2,100. Salesforce and model costs come on top.
A first version scoped to one task with one or two integrations typically takes six to twelve weeks. See our AI agent cost guide and pricing page.
Frequently asked questions
Can we use our own AI model with Salesforce Service Cloud?
Yes. The chatbot runs your model outside Salesforce and reads and writes cases through the REST API. Inside Salesforce, Flow can call your model through External Services, and Apex through HTTP callouts.
What is the Salesforce Agent API for?
It lets an outside application hold a session with an Agentforce agent: start a session, send messages and close it. It brings Salesforce's agent into a channel or app you build; it doesn't run your own model.
Do we need a connected app or an external client app?
An external client app. Salesforce restricted creating new connected apps as of Spring '26, and the Agent API's setup uses an external client app with the client credentials flow.
Does calling an Agentforce agent from our chatbot cost extra?
Yes. Agent API usage draws on Agentforce credits. Salesforce's examples price a 20-credit action at $0.10, or you can buy conversations at $2 each.
How does a rep take over from a custom chatbot?
The chatbot creates or updates a case with a summary of what it already tried, and Omni-Channel routes it to a rep through your existing queues and skills.
Talk to an engineer about your Salesforce chatbot
A conversation about your goals and systems, and a straight answer on whether an agent is the right tool. Paid work starts only after you have agreed a written scope.
- We sign your NDA before the first call if you need one
- You leave with a next step, even if the answer is "you don't need an agent"
- The code and IP are yours if you go ahead
Book a free first call
Pick a time, or send us the details and we will reply.