Bland AI Developer Guide: API, Pricing & Tutorial (2026)

Bland AI is a developer platform for building AI phone agents that make and receive real calls. Its API combines telephony, speech recognition, text-to-speech, conversation logic, tool calls, transfers, testing, and call analytics in one managed service.

For a prototype, you can send a phone number and a natural-language task to one endpoint. For a production workflow, you can use a visual Conversational Pathway, connect your APIs and knowledge base, transfer selected calls to people, and send the outcome back to your application through a webhook.

This Bland AI developer guide shows how the API works, how to make your first call, what the main features cost, and where the platform fits compared with a more modular voice stack.

What Is Bland AI?

Bland AI provides the infrastructure needed to run an automated phone conversation. A developer defines what the agent should accomplish, chooses how tightly to control the dialogue, and connects the agent to the rest of the business.

Typical Bland AI use cases include:

• Qualifying and following up with leads
• Booking, confirming, and rescheduling appointments
• Answering routine customer service calls
• Collecting structured information over the phone
• Looking up orders, accounts, or availability during a call
• Routing a caller to the correct human team
• Running surveys, reminders, and outbound campaigns

The important distinction is that a Bland agent can do more than talk. During the call it can query a knowledge base, call an external API, update a CRM, schedule an appointment, or transfer the conversation. After the call it can return a transcript, summary, extracted variables, disposition, duration, and other metadata.

Bland AI API quick reference

| Item | Details | |---|---| | Primary API | https://us.api.bland.ai/v1 for the US region | | Authentication | API key in the Authorization header; no Bearer prefix | | First endpoint | POST /v1/calls | | Minimum outbound payload | phone_number plus either a task or pathway_id | | Successful response | A unique call_id for status, transcript, and call details | | Live integrations | Tools, webhooks, dynamic data, knowledge bases, and transfers | | Developer tooling | REST API, Bland CLI, MCP server, dashboard, and Web Agent SDK |

Bland AI API Tutorial: Make Your First Call

The shortest path from a new account to a working outbound agent is one API request.

1. Create an account and get an API key

Create a Bland account, open the developer dashboard, and copy your API key. Treat the key like any production secret: keep it in a server-side environment variable and never expose it in browser or mobile code.

For the examples below, store it as BLAND_API_KEY in your local environment.

2. Send a call with cURL

The basic Send Call API needs a destination number in E.164 format and a task. The Authorization value is the API key itself, without the word Bearer.

curl --request POST \
  --url https://us.api.bland.ai/v1/calls \
  --header "Authorization: $BLAND_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "phone_number": "+15551234567",
    "task": "Call the customer to confirm tomorrow’s appointment. If the time no longer works, collect two preferred alternatives. Never invent availability."
  }'

A successful request returns a response like this:

{
  "status": "success",
  "call_id": "9d404c1b-6a23-4426-953a-a52c392ff8f1"
}

The request queues the call; it does not wait for the conversation to finish.

3. Send a call from TypeScript

The same request works with the native fetch API in Node.js:

const response = await fetch("https://us.api.bland.ai/v1/calls", {
  method: "POST",
  headers: {
    Authorization: process.env.BLAND_API_KEY!,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    phone_number: "+15551234567",
    task: [
      "Confirm tomorrow's appointment.",
      "If the time does not work, collect two preferred alternatives.",
      "Do not claim a new time is booked until the scheduling tool confirms it.",
    ].join(" "),
  }),
});
 
if (!response.ok) {
  throw new Error(`Bland API request failed: ${response.status}`);
}
 
const call = (await response.json()) as {
  status: "success" | "error";
  call_id?: string;
  message?: string;
};
 
console.log(call.call_id);

Run this code only on a trusted server. A frontend request would reveal the Bland AI API key to every visitor.

4. Retrieve the call result

Use the returned ID with the Call Details endpoint, or configure a post-call webhook so your application does not have to poll.

curl --request GET \
  --url https://us.api.bland.ai/v1/calls/CALL_ID \
  --header "Authorization: $BLAND_API_KEY"

Call details can include the call status, transcript, summary, duration, recording information, variables, and other result data. Only store the fields your application actually needs, and set retention rules before collecting production calls.

From Demo to Production: Task, Pathway, or Persona?

Bland offers several ways to define an agent. The right choice depends on how much control and reuse the workflow needs.

| Approach | Best for | Main tradeoff | |---|---|---| | Task prompt | A narrow prototype or one-purpose outbound call | Fastest to start, but harder to control as branches multiply | | Conversational Pathway | Multi-step workflows with explicit branches, tools, and failure paths | More setup, but behavior is easier to inspect and test | | Persona | A reusable agent identity and configuration across numbers or use cases | Useful for consistency; individual workflows may still need Pathways |

Use a task for the first proof of concept

A task prompt is ideal when the outcome is narrow: deliver a reminder, confirm one detail, or ask a short list of questions. Write the prompt around an observable outcome, not a character description.

A strong task states:

• Who the agent is calling and why
• What information it may use
• What counts as success
• What it must never guess or change
• When to transfer, reschedule, or end the call
• How to handle voicemail and opt-outs

Once the prompt accumulates many if statements, move the workflow to a Pathway.

Use Conversational Pathways for controlled workflows

Conversational Pathways is Bland's visual, node-based workflow builder. A node can generate dialogue, say a fixed sentence, collect variables, query a knowledge base, call a webhook, transfer the call, or end it. Labeled branches constrain the valid next steps while still allowing the model to understand natural language.

For example, an appointment agent can have separate routes for:

1. Confirming the existing time
2. Requesting a new time
3. Answering a clinic question
4. Handling an unknown caller
5. Escalating to staff
6. Recording an opt-out and ending the call

This structure is easier to review, test, and change than one long system prompt.

Use tools and knowledge for real work

Tools let the voice agent take actions while someone is on the line. Bland provides built-in integrations and custom HTTP tools, so an agent can check availability, retrieve an order, update a CRM, or send data to an internal service.

Knowledge bases provide reference material for questions. Dynamic data supplies per-call context such as a customer name, appointment time, account status, or campaign identifier.

Keep read and write permissions separate. Looking up a booking and canceling it are different authority levels even when they touch the same backend. Validate every input, make write operations idempotent, and require confirmation before an irreversible action.

Webhooks, Transfers, and Call Operations

The call itself is only one part of a production system. Most Bland AI integrations also need a reliable handoff to software or people.

Post-call and live-event webhooks

Set the webhook field on an outbound request to receive the completed call payload. Use webhook_events when you also need events during the call, such as queue, call, latency, tool, or transfer activity.

{
  "phone_number": "+15551234567",
  "pathway_id": "YOUR_PATHWAY_ID",
  "request_data": {
    "customer_name": "Jordan",
    "appointment_date": "September 20, 2026"
  },
  "metadata": {
    "customer_id": "cus_123",
    "campaign": "appointment-confirmation"
  },
  "webhook": "https://example.com/webhooks/bland"
}

The post-call webhook guide explains dashboard and per-call configuration. In production, verify webhook signatures, return a success response quickly, process expensive work asynchronously, and design for retries and duplicate deliveries.

Human transfers

A voice agent should know when to stop automating. Bland can transfer an active call to a single number, route different intents to different teams, or use transfer nodes inside a Pathway.

Treat handoff as a normal workflow state. Define the transfer reason, pass the facts already collected, and decide what happens when no representative is available. A cold transfer without context forces the caller to start over and hides failure from your metrics.

Inbound calls and telephony

For inbound calls, purchase or activate a number and assign an agent in the Bland dashboard. Each number can have its own configuration and webhook behavior. Teams with existing carrier infrastructure can connect Twilio, while enterprise deployments can use SIP integrations.

Phone-number availability, carrier costs, calling regions, transfer behavior, and regulatory obligations vary. Validate the full telephony path in every market where the agent will operate.

Batch calls

Batch Calls can launch an outbound campaign from a CSV or API payload. Per-recipient variables let one Pathway refer to the correct name, appointment, account state, or other context.

Start with a small cohort. Measure answer rate, task completion, opt-outs, transfers, and cost before increasing concurrency. Sending more calls only scales whatever quality and compliance behavior already exists.

Testing and Monitoring Bland AI Agents

A voice agent that works in a scripted demo is not ready for production. Real callers interrupt, stay silent, provide partial information, spell names, reach voicemail, change topics, and ask for a person.

Bland's testing stack includes:

• Testbed: iterate on individual Pathway nodes using real call data
• Scenarios: simulate caller personas and expected situations
• Standards: turn known interactions into repeatable regression checks
• Evals: grade completed calls against configurable quality criteria
• Guard Rails: monitor live responses and trigger an action when a configured requirement fails
• Alerts: notify a team when call metrics or custom conditions cross a threshold

Run each important scenario more than once because model behavior is probabilistic. Cover the happy path, then add silence, interruptions, background noise, accents, numbers, proper nouns, tool timeouts, unavailable staff, and explicit opt-outs.

Guard Rails are an additional control, not a substitute for scoped tool permissions, consent management, representative monitoring, or legal review.

Bland AI CLI and MCP for Developers

The Bland CLI supports a code-oriented workflow for calls, Pathways, phone numbers, logs, local webhook forwarding, validation, testing, and promotion. That makes Pathway changes easier to review and repeat than dashboard-only edits.

Bland also exposes an MCP server for compatible coding agents. Developers can connect tools such as Codex, Cursor, Claude Code, VS Code, and other MCP clients to a Bland account and work with calls or Pathways through structured tools.

MCP is a developer and operator interface to Bland. It is not the real-time voice model speaking to the person on the phone.

Bland AI Pricing in 2026

Bland combines a platform tier with usage charges. The public prices checked on September 19, 2026 were:

| Plan | Platform fee | Connected-minute rate | Transfer-minute rate | Published limits | |---|---:|---:|---:|---| | Start | $0 | $0.14 | $0.05 | 10 concurrent calls, 100 calls/day, 10 knowledge bases | | Build | $299/month | $0.12 | $0.04 | 50 concurrent calls, 2,000 calls/day, 50 knowledge bases | | Enterprise | Custom | Custom | Custom | Concurrency and volume sized to the deployment |

The Start plan includes two initial credits and an inbound number, but it is not unlimited free calling. Bland bills connected time to the second. The connected-minute rate includes the LLM, speech-to-text, and text-to-speech layers; carrier and telephone-number charges are separate.

A useful cost model is:

monthly cost ≈ platform fee
             + connected minutes × talk-time rate
             + transfer minutes × transfer rate
             + carrier, phone-number, and optional feature charges

The lower Build rate alone offsets its $299 platform fee at roughly 14,950 connected minutes per month. In practice, teams may need Build earlier for its higher daily and concurrency limits. Use your own answer rate, average call length, transfer rate, and failed-attempt volume when forecasting.

Bland can use its built-in carrier setup at pass-through cost, connect to a customer's Twilio account, or support SIP configurations for enterprise deployments. Bring-your-own-telephony customers do not pay Bland's transfer fee but still pay their carrier and Bland's AI usage rate.

Enterprise adds options such as dedicated infrastructure, on-premises or VPC deployment, SSO, data residency, BAA support, higher limits, and forward-deployed engineering. Check the official Bland AI pricing page before buying because rates, limits, and plan availability can change.

The separate Agent Phone Plan

Bland also sells a $29.99 monthly Agent Phone Plan for one autonomous AI agent using one US number for US and Canada voice and SMS, subject to fair-use limits. It allows one active call at a time and is separate from the developer plans above. It is useful when an assistant needs its own line, not as the pricing model for a high-volume contact center.

Bland AI vs. Vapi, Retell AI, and a Custom Stack

The main decision is not simply which demo sounds best. It is how much of the voice stack your team wants to assemble and operate.

| Option | Best for | What you operate | Main tradeoff | |---|---|---|---| | Bland AI | Teams that want one managed voice platform, visual workflows, and bundled AI usage | Business logic, integrations, prompts, and monitoring | Less component-level choice than a modular stack | | Vapi | Developers who want to compose model, voice, and telephony providers | More provider configuration and cost layers | Greater flexibility with more integration decisions | | Retell AI | Teams comparing another integrated voice-agent API and builder | Agent configuration, integrations, and operations | Different workflow, feature, and pricing model to evaluate | | Custom or open-source stack | Teams with unusual model, latency, regional, or infrastructure requirements | Telephony, orchestration, speech, models, scaling, and observability | Maximum control with the highest engineering burden |

Bland is strongest when speed to production, predictable AI-layer billing, and one operational surface matter more than choosing every underlying model. A modular stack is more attractive when provider portability or fine-grained infrastructure control is a core requirement.

Bland AI Pros and Cons

Advantages

• One API covers inbound and outbound voice-agent workflows
• The basic call API is fast to prototype with
• Pathways provide more control than a single long prompt
• Tools, knowledge, transfers, batches, and webhooks are integrated
• Testing, evals, guard rails, and alerts support production operations
• CLI and MCP options fit code-first teams
• Twilio, SIP, and enterprise deployment options cover more complex telephony needs
• LLM, speech-to-text, and text-to-speech usage is bundled into the talk-time rate

Limitations

• Pathways and prompts are specific to Bland, which increases switching work
• A bundled stack gives less freedom to swap every model or speech provider
• Headline minute rates do not include every carrier, number, transfer, or optional cost
• Several governance and advanced routing features are enterprise-only
• Voice quality and latency still need testing with your callers, languages, tools, and phone routes

When Bland AI Is a Good Fit

Choose Bland AI when phone calls are a core workflow, you need inbound and outbound agents on one platform, and conversations require live data, actions, branching, or human handoff. It is especially practical when an operations team wants a visual builder while developers retain API and CLI access.

It may be more platform than you need for a single low-volume notification. It may also be the wrong fit if your architecture requires complete control over the language model, speech vendors, or open-source runtime.

Production Checklist

Before increasing call volume:

1. Define one measurable outcome. A call should confirm, qualify, book, collect, route, or resolve something observable.
2. Constrain authority. Give the agent only the data and actions required for that outcome.
3. Design the human handoff. Transfer context, define after-hours behavior, and measure unsuccessful transfers.
4. Handle consent and opt-outs. Validate disclosure, recording, calling-hour, and do-not-call requirements with qualified counsel for every jurisdiction.
5. Secure integrations. Keep secrets server-side, verify webhook signatures, authenticate tools, and make writes idempotent.
6. Test difficult conversations. Include silence, interruptions, accents, voicemail, background noise, tool failures, and off-script requests.
7. Measure the full funnel. Track answer rate, completion, escalation, opt-out, latency, retries, and total cost per successful outcome.
8. Roll out gradually. Start with internal calls, then a small representative cohort, and raise limits only when the data supports it.

Bland AI FAQ

Does Bland AI have an API?

Yes. Bland exposes a REST API for calls, Pathways, numbers, tools, knowledge bases, batches, testing, logs, and other resources. The basic outbound endpoint is POST /v1/calls.

Where do I get a Bland AI API key?

Create or sign in to a Bland account and copy the key from the developer dashboard. Send it in the Authorization header without a Bearer prefix, and keep it in server-side secret storage.

Can Bland AI receive inbound calls?

Yes. Activate or import a phone number, attach an agent, and configure the number's Pathway, prompt, webhook, and other behavior. Outbound calls use the same platform through the API, dashboard, or batches.

How much does Bland AI cost?

The Start plan has no platform fee and charges $0.14 per connected minute. Build costs $299 per month plus $0.12 per connected minute. Transfers, carrier usage, numbers, and optional services can add cost. Enterprise pricing is custom.

Is Bland AI free?

Bland is free to start, not free for ongoing usage. The Start plan has a $0 platform fee and includes initial credits, after which connected calls are usage-based.

Does Bland AI support HIPAA workloads?

Bland states that HIPAA eligibility with a signed BAA is available on Enterprise. A vendor feature does not make an application compliant by itself; review data flow, access, retention, recording, and operational controls with appropriate legal and security teams.

Can I connect Bland AI to Codex or another coding agent?

Yes. Bland documents an MCP server and client setup for Codex, Cursor, Claude Code, VS Code, and other compatible tools. The MCP connection can help a coding agent manage Bland resources, while the voice agent handles the live phone conversation.

Bottom Line

Bland AI turns a complicated real-time voice stack into a direct developer surface: define the conversation, connect business actions, attach a number, and call the API. A simple task can validate an idea quickly; Pathways, tools, webhooks, tests, and transfers provide the controls needed for a real workflow.

The tradeoff is the same as with any managed platform: you gain speed, integrated tooling, and simpler AI-layer billing, while accepting Bland's workflow model, pricing, and place in your critical path. If those constraints fit, Bland AI is a strong option for production phone agents without owning the entire voice infrastructure.