Veeva added artificial intelligence to its applications under the names Veeva AI and Vault AI, and it positioned the Direct Data API as the way to pull Vault data into customers’ own AI and analytics tools. The licensing facts that Veeva makes public are few. They are the basis on which the AI is licensed, the meters an administrator can see, and the technical limits on the API. This article describes those facts, and notes where the public sources are silent. Veeva states that product terms are governed solely by the written customer agreements, so these statements describe mechanics and intentions, not entitlements.[11]
Editions
Veeva AI and Vault AI
When Veeva announced Veeva AI on 2025-04-29, it described AI Agents and AI Shortcuts added to the Vault Platform and to Veeva applications, and said that Veeva AI is LLM agnostic. Customers may use a Veeva-supplied LLM or configure Veeva AI to use a customer-specific LLM.[1] The same release states the licensing basis: the first release “will be licensed at the Vault level with a simple and reasonable subscription fee”.[1] A Vault-level licence is a different unit from the per-user licences described in Vault license types and application licensing: it attaches to the Vault, not to named users. Whether Vault AI is enabled per Vault, for all applications in a Vault or per application is not stated in the sources. Catalog rows: Veeva AI; Veeva AI is licensed at the Vault level; Veeva AI supports Veeva-supplied or customer LLMs.
The Vault AI product page says that Vault AI Agents use large language models from Anthropic and Amazon, hosted on Amazon Bedrock, and that custom agents use Veeva-hosted models or customer-provided models hosted on Amazon Bedrock or Microsoft Azure AI Foundry.[2] The Vault Platform page adds that agents can be configured or created by customers.[10]
Falcon
Veeva Falcon is described as agentic labor: standard agents that execute multi-step processes in clinical, regulatory, safety and commercial work, with Falcon MLR available and the first clinical, regulatory and safety agents planned for November 2026.[3] The page does not state a licence basis for Falcon. It should not be assumed to share the Vault-level basis announced for Veeva AI.
Direct Data API
On 2025-02-27 Veeva announced that the Direct Data API “is now included for no additional license fee as part of Veeva Vault Platform”.[6] The product page describes a new class of API that makes Vault data accessible up to 100 times faster than traditional APIs, with open-source accelerators for Amazon Redshift, Snowflake, Databricks and Microsoft Fabric.[7] Catalog rows: Direct Data API; Direct Data API included in Vault Platform. Link data has its own Direct Data API, described in Veeva data products licensing.
Metrics
LLM tokens
Token monitoring is available “only to customers who have Vault AI enabled on their Vault”.[4] In Admin > Settings > Vault AI Settings an administrator can enter a maximum 30-day token usage limit in the millions, and can enter an email address for alerts.[4] The limit works as an alert threshold. Vault sends weekly email alerts at 50, 75 and 85 percent of the limit, a daily alert above 95 percent, an hourly alert when use exceeds five times the limit, and alerts on spikes in daily use.[4] The help page does not say that exceeding the limit blocks use, so the limit is best read as a customer-configured control, not a licensed cap. Whether Veeva bills on token use is not stated in the sources.
The 30-day figure counts tokens from both Veeva-provided LLM connections (Vault AI Basic and Vault AI Advanced) and custom LLM connections, so it measures all AI use in the Vault, whoever hosts the model.[4] Standard and system agents use Vault AI Basic and Vault AI Advanced respectively, and a customer can select its own configured LLM for custom agents.[5] For a custom LLM connection the administrator sets the maximum output tokens, between 512 and 10,000, with a default of 4,096.[5] Catalog rows: Vault AI LLM tokens (30-day limit); Vault AI functionality needs Vault AI enabled; 30-day LLM token limit is set by the admin; Token alert thresholds; Veeva-provided and custom LLM connections.
Agent activity records
Vault records each agent action in Agent Instance and Agent Action Execution objects, with input, output and total tokens per execution. Those two record types older than 60 days are deleted weekly, while Daily Agent Activity and Daily Agent Action Activity summaries are never deleted.[4] A customer that wants to reconstruct AI usage for a past period therefore needs the daily summaries, which record tokens by LLM connection and so can separate Veeva-provided from custom connections.[4] The Vault AI Agent Activities dashboard shows the past 60 days.[4] Agent users are System Managed Users, which are not included in license counts.[13]
API burst limit
The Vault API limits the number of calls in a fixed five-minute period. When the burst limit is reached, the server delays responses for the remainder of the period, and the example in the documentation is 2,000 requests in 5 minutes.[8] Vault Help states that after the burst limit each subsequent API call is delayed by 500 ms until the period ends.[9] Authentication calls have a separate one-minute burst limit, tracked by user name and Vault DNS, which does not apply to SAML/SSO or OAuth, and when it is reached further authentication requests fail.[8] As of v21.1, Vault does not enforce daily API limits and does not notify users when API transaction limits are partly reached.[8] Catalog rows: Vault API burst limit; Vault API burst limit; Daily API limits are not enforced.
Admins can see the burst API count under Admin > About > Vault Information, and integration developers can add an optional client ID to API calls, which can be enforced through client ID filtering on external connections.[9] Veeva describes the client ID as a way to understand where an API call originated.
Counting and floors
- Vault AI is not a user count. Veeva announced a Vault-level basis, and administrators control token alerts, so the number of people using AI does not itself appear as a licence metric in the public documents.[1][4]
- Agents are not licensed users. Agent Users are System Managed Users, and system-owned users operate with Full User licenses but are not included in license counts.[12][13]
- No stated floors. No minimum fee, token allotment or API quota is stated for any of these items.
Virtualization and partitioning
AI settings, token limits and agent records are Vault-level settings and records. A sandbox Vault created or refreshed from a snapshot sets agent monitoring records to inactive, and such inactive records are typically deleted the day after the sandbox is created or refreshed.[4] AI use in a sandbox is therefore not preserved in the same way as in production. Whether sandbox AI use is licensed separately is not stated.
Cloud and BYOL
Customers may bring their own LLM in the sense that Veeva AI can be configured with a customer-specific LLM, and custom agents can use customer-provided models hosted on Amazon Bedrock or Microsoft Azure AI Foundry.[1][2] The licence for such a model is the customer’s own arrangement with that provider, which Veeva does not describe. Tokens consumed by it still appear in the Vault’s 30-day token usage figure.[4]
Programs
- Veeva AI Partner Program. Veeva says it supports customers and partners to develop AI solutions using the Direct Data API and the Veeva AI Partner Program.[6] The terms of the program are not published in the pages cited.
- Direct Data API inclusion. The no-additional-fee inclusion is the one documented commercial concession.
Out of scope
This article does not cover the accuracy or governance of AI agents, the contents of Veeva’s AI terms, or the licences of third-party LLM providers. It does not cover Veeva Network or the data products.