Pipedrive · Solution guide
Pipedrive AI Agents: Built-In Pipedrive AI vs a Custom Agent on the API (2026)
Written scope and quote after the call. NDA on request. You own the code.
Quick answer
What Pipedrive's own AI does today (Sales Assistant, AI email creation, AI notifications, Nova meeting intelligence, MCP access for ChatGPT and Claude), which plans include each feature, and when a business needs a custom Pipedrive AI agent built on the API and webhooks instead. Includes the jobs a custom agent does, the records it reads and writes, how we build it with human approval and token budgets, and what it costs to run.
Builds like this
Part of the guide: CRM AI agents
Key takeaways
- Pipedrive's built-in AI, per its Pipedrive AI support article checked Oct 31, 2026, six of its ten listed features are on all four plans: the AI import assistant, AI report creation, AI Marketplace search, smart app recommendations, the mobile business card scanner and Sales Assistant (beta). The other four, AI email creation, AI notifications, email summarization (beta) and suggested replies (beta), need Premium ($59 a seat a month, billed yearly) or Ultimate ($79).
- Nova, Pipedrive's meeting intelligence tool, is included in all plans at no extra cost. It drafts CRM updates after a call, and nothing reaches the pipeline until a person approves it.
- Pipedrive MCP lets ChatGPT, Claude or another MCP client read and update Pipedrive records on every plan, within the connected user's permissions, with actions recorded in the change log. Usage draws on plan token limits.
- We found no Pipedrive product that runs on its own in the background, such as qualifying every new lead or chasing stalled deals overnight. That is the gap a custom Pipedrive AI agent fills: it listens to webhooks, reads records through the API, decides with a language model and writes back under rules you set.
- A custom agent spends the same daily API token budget as every other integration: 30,000 tokens x plan multiplier (Lite 1, Growth 2, Premium 5, Ultimate 7) x seats. A search costs 40 tokens and an update 10, so the agent's design is also a budget decision.
A Pipedrive AI agent is software that watches your Pipedrive account, reads the deal, person, organization, lead and activity records involved, decides what to do with a language model, and writes the result back: a lead score, a note, a next activity, a stage move or a drafted email for a rep to approve. Pipedrive ships useful AI of its own, and for many teams that is enough. This page sets out exactly what Pipedrive’s built-in AI does and on which plan, where it stops, and what a custom agent built on the Pipedrive API and webhooks adds. Every Pipedrive fact below comes from pipedrive.com or support.pipedrive.com, checked Oct 31, 2026.
Sources: Pipedrive AI support article, pricing page, Nova page and rate limiting docs, all checked Oct 31, 2026.
What AI is built into Pipedrive, and on which plan?
Pipedrive’s own availability table lists ten features (Marketplace search and app recommendations share one row below, so they take nine rows), plus Nova and MCP, which have their own pages. The split that matters is between assistants that help one person do a task faster and anything that changes CRM data, because every write-capable feature keeps a person in the loop.
| Feature | Lite ($14) | Growth ($39) | Premium ($59) | Ultimate ($79) | What it does |
|---|---|---|---|---|---|
| AI import assistant | Yes | Yes | Yes | Yes | Maps spreadsheet columns to Pipedrive fields during import |
| AI-assisted report creation | Yes | Yes | Yes | Yes | Builds an Insights report from a typed prompt |
| AI Marketplace search and app recommendations | Yes | Yes | Yes | Yes | Finds and suggests Marketplace apps |
| Business card scanner (mobile) | Yes | Yes | Yes | Yes | Turns a card photo into a contact |
| Sales Assistant (beta) | Yes | Yes | Yes | Yes | Answers questions about your deals, activities and contacts, drafts notes and emails, explains features |
| AI email creation | No | No | Yes | Yes | Drafts an email from a prompt, tone and length in Sales Inbox or a detail view |
| AI-powered notifications | No | No | Yes | Yes | Win-probability style tips and alerts under Analytics and tips |
| AI email summarization (beta) | No | No | Yes | Yes | Summary, sentiment, readiness to buy (1 to 10) and action items for a thread |
| AI suggested replies (beta) | No | No | Yes | Yes | Drafts a reply from the email thread |
| Nova meeting intelligence | Yes | Yes | Yes | Yes | Pre-call brief, recording, transcript, recap and drafted CRM updates you approve |
| Pipedrive MCP | Yes | Yes | Yes | Yes | Lets ChatGPT, Claude or another MCP client query and update records |
Sources: Pipedrive AI support article for the first nine rows (ten features), Nova and MCP pages for the last two, pricing for per-seat monthly prices billed yearly, all checked Oct 31, 2026. Beta features are, in Pipedrive’s words, “available only to selected users” through its beta program, and an account settings admin has to enable AI features for the company first.
A few details from the support pages are worth knowing before you plan around them. Sales Assistant is powered by OpenAI, shows the sources behind each answer, keeps each user’s chat history private even from admins, and its support article says your data is never used to train AI models. AI email creation produces a draft you copy into the message, so nothing is sent without a rep. Nova’s page says you can “approve, edit or skip every CRM update before it gets to your pipeline”, and it puts consent for recording on the meeting organizer. For the dated release history of Nova and the MCP admin switch, see our Pipedrive product updates 2026 page.
Does Pipedrive have an AI agent that works on its own?
Not as of Oct 31, 2026, as far as Pipedrive’s own pages show. Everything above is either a person asking an assistant for help or a draft waiting for a person to approve it. Pipedrive MCP comes closest: once connected, an assistant such as Claude or ChatGPT can create a deal from meeting notes, move a deal to a stage and schedule a follow-up. But it acts when a user types a request, within that user’s permissions, and Pipedrive records the actions in the change log.
What none of these do is run a job on every new record without being asked. Examples of jobs that need something that runs by itself:
- Every new lead from a web form gets researched, scored against your ideal customer rules and routed to an owner before a rep opens it.
- Every deal with no activity for 10 days gets a drafted follow-up and a task for its owner.
- Every inbound reply is classified (interested, not now, wrong person, unsubscribe) and the lead or deal is updated to match.
- Every won deal gets a handoff summary written into a note for the delivery team.
Pipedrive’s own workflow automations can do the rule-based part of these jobs, such as “if stage changes, create activity”. The judgment part, reading an email or a website and deciding what it means, is what the language model in a custom agent adds.
When is Pipedrive’s built-in AI enough, and when do you need a custom agent?
Start with the built-in features. They cost nothing extra on your plan, they already respect Pipedrive’s permissions, and they need no code. Move to a custom agent only when a specific job is not covered.
| Situation | Built-in Pipedrive AI | Custom Pipedrive AI agent |
|---|---|---|
| Reps want help writing emails | AI email creation and suggested replies on Premium and Ultimate | Not needed |
| Managers want answers about the pipeline | Sales Assistant, AI reports | Not needed |
| Meeting notes should update deals | Nova, with approval | Only if notes come from a source Nova does not cover and transcript upload does not fit |
| Founders want to update Pipedrive from ChatGPT or Claude | Pipedrive MCP | Not needed |
| Every new lead should be scored and routed without anyone asking | Not covered | Yes |
| Inbound replies should be classified and records updated automatically | Not covered | Yes |
| The agent must follow your own rules, pricing or playbook | Partly, through prompts each time | Yes, rules live in the agent |
| Data must combine Pipedrive with another system, such as billing or a product database | Not covered | Yes |
| You are on Lite or Growth and need AI email drafting | Upgrade to Premium | A custom agent can draft, but compare its build cost with the upgrade first |
Source: aibrevo scoping guidance, Oct 31, 2026, against the Pipedrive feature list above. The last row matters: if the only gap is AI email drafting, upgrading a few seats to Premium is usually simpler than a build. Use our Pipedrive pricing calculator to price the upgrade for your seat count.
Jobs a custom Pipedrive AI agent does in the CRM
It does a defined job, triggered by a defined event, reading and writing defined records. We write each agent down in exactly those terms before building it, because “an AI that helps sales” cannot be tested and a job like the ones below can.
| Job | Trigger | Reads | Writes | Human approval |
|---|---|---|---|---|
| Lead qualification | Webhook on a new lead or person | Lead, person, organization, form fields, the company website | Score custom field, qualification note, owner, label | Optional, usually only for leads above or below a threshold |
| Stalled deal follow-up | Scheduled check for deals with no activity for N days | Deal, last activities, last emails | Drafted email in a note, follow-up activity for the owner | Yes, the rep sends the email |
| Reply triage | Inbound email synced to a deal or lead | Email body, deal stage, person | Label, next activity, stage change for clear cases | Yes for stage changes, no for labels |
| Won-deal handoff | Webhook on deal won | Deal, products, notes, activities, files list | Handoff summary note, project or onboarding task | No, it only writes a note |
| Data hygiene | Nightly batch | Persons and organizations changed that day | Normalized phone and country fields, flags for likely duplicates | Yes for merges |
Source: aibrevo agent design patterns for Pipedrive, Oct 31, 2026. These are examples of how we specify agents, not client results.
Two Pipedrive mechanics shape every one of these. Custom fields are addressed in the API by a generated key, not by their label, so the agent reads the field list rather than hard-coding keys that break when someone recreates a field; our Pipedrive API integration guide explains the trap in detail. And Pipedrive’s merge behavior sends a delete event with a merged_to_id value, per its webhooks v2 guide, so an agent that keeps its own memory of records has to follow merges instead of treating them as deletions.
How do we build a Pipedrive AI agent?
Our in-house engineering team, 16 engineers led by co-founder and CTO Shivam, builds agents on Pipedrive in six steps. Each step ends with something you can check.
1. Job and rules
One job per agent, its trigger, the fields it may write and the cases where it must ask a person. You can check: a one-page agent spec.
2. Access and permissions
An OAuth app or a dedicated API user with only the visibility the job needs, so the agent cannot see pipelines it has no reason to touch. You can check: the scopes and the user's permission set.
3. Events
Webhooks v2 subscriptions such as create.lead or change.deal, protected with HTTP Basic Auth, acknowledged inside Pipedrive's 10-second timeout and processed from a queue. You can check: the subscription list and a delivery log.
4. Build and prompts
The model call, the rules around it and the write-back, with every proposed change stored before it is applied. You can check: sample outputs on your own anonymized records.
5. Test on real records
A shadow run where the agent only writes its decisions to a log, compared with what your reps did. Ultimate plans include a sandbox account for this. You can check: an agreement rate and a list of misses.
6. Go-live and guardrails
Writes switched on job by job, with a kill switch, a token ceiling and a weekly review of a sample. You can check: the dashboard and the first review.
The guardrails we put on every agent:
- Human approval where a mistake reaches a customer. Emails are drafted, not sent, unless you decide otherwise for a narrow case. Stage changes need approval until the agent has a measured record.
- Logging of every decision. Input record IDs, the model’s answer, what was written and the API response, so any change in Pipedrive can be traced to a reason.
- Token budgeting. Pipedrive’s budget is 30,000 tokens x plan multiplier x seats per day, shared by every integration on the account, and once it is spent the API returns 429 until the reset at midnight in the server’s time zone. Pipedrive emails admins at 75% and 100%. Our agents stop themselves well before that point so your other integrations keep working.
- Burst control. Pipedrive also limits each token on a rolling 2-second window, for example 40 requests for an API token or 160 for an OAuth app on Growth, so writes go through a queue with a concurrency cap.
- Loop prevention. An agent that updates a deal fires a
change.dealwebhook. We tag the agent’s own writes and ignore the events they cause, otherwise the agent can spend the day’s budget responding to itself. - Personal data. Only the fields the job needs go to the model, the model provider’s data terms are agreed with you, and special-category or health data stays out unless your compliance review says otherwise. If health data is involved, read is Pipedrive HIPAA compliant first.
- An MCP decision. If your company does not want staff connecting their own AI assistants to Pipedrive, a global admin can switch MCP off in Company Settings without affecting a custom agent that uses an API token or OAuth app.
Webhook facts in steps 3 and 5 are from Pipedrive’s webhooks v2 guide, checked Oct 31, 2026: failed deliveries are retried after 3, 30 and 150 seconds, 10 first-attempt failures ban a webhook for 30 minutes, a webhook with no successful delivery for 3 days is deleted, and each user can hold up to 40 webhooks.
The API budget a Pipedrive AI agent uses
Plan the agent’s token use before building it, because it shares one pool with your other apps. Pipedrive lists these costs for some operations: get a single entity 2 tokens, get a list 20, update a single entity 10, delete 6, search 40 (rate limiting docs, checked Oct 31, 2026). It also says v2 endpoints cost fewer tokens than their v1 equivalents.
Chart: 30,000 base tokens x plan multiplier x 10 seats, before top-ups. Source: Pipedrive rate limiting documentation, checked Oct 31, 2026.
An illustrative lead-qualification agent that does one search for duplicates (40), reads the person and organization (2 + 2) and writes one update (10) spends about 54 tokens per lead. On a 10-seat Growth account that is roughly 1% of the daily budget per 110 leads, so lead volume is rarely the problem. Searches inside loops are: an agent that searches for every deal on every stage change can spend the budget quickly. If you run short, Pipedrive sells API token top-ups of 250,000 tokens, up to 10 per company, from the Growth plan (top-ups article, checked Oct 31, 2026; prices shown at checkout). How Pipedrive’s budget compares with monday.com and GoHighLevel is in our CRM API rate limits comparison.
What does a Pipedrive AI agent cost to run?
Four running costs, plus the one-time build:
| Cost | Who bills it | What sets it |
|---|---|---|
| Pipedrive seats | Pipedrive | $14, $39, $59 or $79 per seat a month billed yearly (pricing page, checked Oct 31, 2026); prices exclude VAT |
| API token top-ups | Pipedrive | Only if the agent and your other integrations exceed the daily budget |
| Language model usage | The model provider, on your account | Number of records processed and how much text each one sends |
| Hosting, queue and logs | Your cloud account or ours, agreed in scope | Usually small next to the other three |
| Build and changes | aibrevo | Scoped and quoted per agent after the free call |
We do not publish a price for Pipedrive agent builds, because the cost depends on the number of jobs, the systems involved and how much approval logic you want. Each project is scoped and quoted after a free 30-minute call. If you also need the Pipedrive account itself set up or cleaned first, that is our Pipedrive implementation service.
How long does a custom Pipedrive AI agent take?
Our usual scoping range for a single-job agent that writes notes, scores or labels is two to four weeks from spec to go-live, and much of that is the shadow run on your real records. An agent that drafts customer emails, changes stages or combines Pipedrive with another system is typically four to eight weeks, because each extra write path needs its own approval rule and test. What slows projects most is waiting for agreement on the rules (“what counts as a qualified lead?”), not the code.
Checks before an AI agent writes to Pipedrive
- Field ownership. If a sync from another tool also writes the same field, the agent and the sync will overwrite each other. Name one owner per field.
- Automations already running. A Pipedrive automation on “deal updated” will fire on the agent’s writes too. List them before go-live.
- Who the agent acts as. Writes appear in the change log under the API user or app, so use a clearly named one.
- Permissions drift. If the agent’s user is given wider visibility “temporarily”, it stays wider. Review it with each change.
- Model drift. A model update can change answers. Keep a fixed test set of records and rerun it when the model or prompt changes.
- Recording consent. If the agent uses Nova transcripts or call recordings, the organizer’s consent duties that Pipedrive describes on its Nova page still apply.
Who is a custom Pipedrive AI agent for, and who is it not for?
It suits Pipedrive teams with a repeating, rule-heavy job that people are doing by hand: a steady flow of inbound leads, many open deals per rep, or a handoff from sales to delivery that keeps losing detail. It suits teams that already have clean stages and owners, because an agent automates the process you have.
It is not for a team whose pipeline stages are still changing every month, a team that only needs AI email drafting (upgrade to Premium instead), or a team that has not yet set up basic Pipedrive automations. In those cases we will say so on the call. If you are still choosing a CRM, our Pipedrive guide covers plans and limits first.
Want a Pipedrive AI agent scoped? Book a free 30-minute call and bring one job you want handled, your Pipedrive plan and seat count, and the other tools involved. You get a written agent spec with the trigger, the records it reads and writes, the approval rules and a quote. Prefer writing? Send a message.
Sources
All pages opened on Oct 31, 2026 through a reader proxy, because pipedrive.com blocks automated requests.
- Pipedrive support, Pipedrive AI: AI feature availability by plan, enabling AI features, beta notes.
- Pipedrive support, Sales Assistant: capabilities, OpenAI, sources tab, chat history privacy, data not used for training.
- Pipedrive support, AI email creation: Premium and higher only.
- Pipedrive, Pipedrive AI and AI Sales Assistant: feature descriptions.
- Pipedrive, Nova: included in all plans, approve or skip CRM updates, consent responsibilities.
- Pipedrive, Pipedrive MCP: official Claude and ChatGPT connector, permissions, change log, all plans, token limits.
- Pipedrive, Pricing: Lite $14, Growth $39, Premium $59, Ultimate $79 per seat a month billed yearly; sandbox on Ultimate.
- Pipedrive developers, Rate limiting: token budget formula, endpoint costs, burst limits, 75% and 100% emails.
- Pipedrive developers, Guide for Webhooks v2: Basic Auth, 10-second timeout, retries, ban system, 40 webhooks per user,
merged_to_id. - Pipedrive support, Top-ups: API token top-ups of 250,000, up to 10, from Growth.
