The Dataiku LLM Mesh is the gateway layer through which Dataiku applications connect to large language model services. Dataiku’s product page describes it as centralising how teams connect to, manage and govern LLM services, with routing, screening, moderation, cost tracking, auditing, caching of frequent responses and usage limits.[1] Licensing questions around it fall into three groups: what is included with the platform, what requires the Advanced LLM Mesh add-on, and how the optional AI Services (assistants such as Cobuild) are metered under their own terms.
The documentation does not publish prices or part numbers for any of these items. What it does publish are gates (a feature that needs the add-on), metering rules (included usage and credits) and data handling rules, which are the elements an asset manager can verify against a customer’s Order.
Editions
What requires Advanced LLM Mesh
The Cost Control documentation states that a licence with Advanced LLM Mesh is needed to customise quotas; without it, only a single global quota (the Fallback Quota) can be set, applied to all LLM queries.[2] The RAG guardrails documentation states that RAG guardrails, which set minimum faithfulness and relevancy thresholds on retrieval-augmented models, are available to customers with the Advanced LLM Mesh add-on.[3] The DSS 13 release notes label three capabilities as part of the add-on: LLM cost blocking and alerting, RAG guardrails and visual fine-tuning of models. They also mention features offered in a private preview as part of an Advanced LLM Mesh Early Adopter Program.[4] See Advanced LLM Mesh add-on.
The 14-day Dataiku Cloud trial includes all features except Govern and advanced LLM Mesh, so trial testing does not show how the add-on behaves.[11]
Metrics
LLM Mesh cost quota
A quota in Cost Control tracks LLM spend in US dollars. For each quota the administrator sets the scope (all queries, or a filter by provider, project, connection or user), the quota amount, whether queries are blocked when it is reached, the reset period (for example every calendar month or a rolling 15 days), optional email alert thresholds and visibility permissions.[2] A single query can match several quotas, and every matching quota is incremented.[2] Cost tracking is not supported for Amazon SageMaker LLM, Databricks Mosaic AI or Snowflake Cortex, so quotas cannot target those models specifically, though their queries can still be blocked if a matching blocking quota is exhausted; local Hugging Face models cannot be targeted or blocked by quotas.[2] Only administrators create, update or delete quotas, and they can grant full or basic visibility to other users or groups.[2] See LLM Mesh cost quota.
The quota is a spending guard and is distinct from what Dataiku charges. It measures the customer’s spend with LLM providers; the customer pays those providers under its own agreements, because the License Terms state that use of third-party platforms is subject to the customer’s agreement with the provider and not to the License Terms.[9]
AI Credits and Monthly Included Usage
AI Services use a credits-based model. Every paid Designer seat benefits from Monthly Included Usage that is specific to each Designer, is not poolable across developers, and replenishes on the first calendar day of each month. Usage can be extended through the purchase of Dataiku AI Credits, which customers buy through their Dataiku Account Executive or Customer Success Manager.[6] After a user’s Monthly Included Usage is spent, usage consumes credits; if the customer has no credits left, Cobuild and other AI Services are blocked for that user until Monthly Included Usage regenerates on the first day of the next month.[6] Consumption depends on the action, the prompt and the size of the environment, and the documentation gives indicative ranges: 0.1 to 5 credits for generating SQL with SQL Assistant, 0.1 to 10 for explaining a Flow, 1 to 10 for Cobuild help with part of a Flow, and 5 to 50 for Cobuild building a Flow from input data, noting that figures may vary a lot.[6] See AI Credit and Monthly Included Usage.
Counting and floors
Administrators monitor AI Services usage under Admin, AI Services, which shows for each user how much of the Monthly Included Usage and the credits has been consumed and the remaining credit balance.[6] The billing system does not know actual logins, only a hash of them, so entries labelled “Unknown user” correspond to usage by users who exist only on other Dataiku instances.[6] The monitoring view shows each user’s own consumption and the remaining credit balance of the customer.[6]
Dataiku can change the included allowance. The AI Services Terms of Use state that use may include a limited amount of included free usage and/or additional credits subject to fees in the applicable Order, and that Dataiku may modify, suspend or discontinue included free usage at any time.[7] Any fees are set in the Order.[7]
Virtualization and partitioning
The AI Services and LLM Mesh are not metered by processors, cores or virtual machines in the documents reviewed. They are metered per paid Designer seat (included usage), per credit consumed and, for cost control, per US dollar of LLM spend.[6] AI Services also need the Dataiku DSS server to be connected to the internet in both operating modes, with a specific gateway URL to allow through firewalls.[5]
Cloud and BYOL
Two operating modes
The AI Services can run in two modes. In “Operated by Dataiku” (Dataiku-Service) mode, processing happens on Dataiku’s servers, the input is processed by Dataiku and by third-party AI providers that Dataiku selects and manages, and the output is at least in part generated by those providers.[7] In “Bring your own LLM” mode, the customer connects the AI Services to its own LLM through an LLM Mesh connection; the input is not sent to Dataiku and Dataiku does not process input or generate output.[7] The Stories AI service is not available in the bring-your-own-LLM mode.[5] The current list of third-party AI providers used in Dataiku-Service mode is kept in a documentation page.[8]
The AI Services named in the documentation are Cobuild, SQL Assistant, AI Search, Generate Metadata, Stories AI, AI Explain and Generate Steps. SQL Assistant, AI Search, Generate Metadata, AI Explain and Generate Steps are collectively called “Classic AI Assistants”.[5] Separate “Third-Party Assistants” such as coding agents integrated in Code Studios are entirely under customer control and use LLM connections the customer configures, so they do not go through Dataiku’s AI Services.[5]
Terms, data and ownership
Use of AI Services is optional and requires acceptance of the AI Services Terms of Use, linked from the AI Services page in Admin, before the services are turned on.[5] The License Terms and the Cloud Terms both incorporate those terms and state that Dataiku will not use customer data to train AI models.[9][10] The Terms of Use add that Dataiku will not allow any third-party AI provider to use the data to train its models.[7] To the extent permitted by law, Input and Output used with Dataiku-Service mode are the customer’s data, and the customer is responsible for their content and accuracy.[7] Users must not submit special categories of personal data, patient or health information, PCI DSS data or data relating to minors to the AI Services.[7] The customer may not use the AI Services to develop a capability that competes with them.[7]
The warranty is limited. Dataiku warrants that the AI Services perform materially as described in the documentation, but does not warrant the accuracy of any output, and the customer is solely responsible for reviewing and testing output; in bring-your-own-LLM mode the warranties apply only to the components Dataiku provides.[7]
Programs
- Included usage. Monthly Included Usage per paid Designer seat, described above.[6]
- Early adopter programs. Some Advanced LLM Mesh capabilities have been released in private preview under an Early Adopter Program, according to the release notes.[4]
- Optional and separate offering. The Terms of Use describe the AI Services as an additional and optional offering; a breach of the AI warranty is remedied exclusively under the AI terms and does not affect the rest of the agreement or other warranties and service levels.[7] Dataiku’s FAQ for self-managed software states that the optional AI Services can be used through a Dataiku-operated service mode or through the customer’s own LLM.[12]
Audits and compliance
Dataiku collects AI Services usage information to monitor consumption, and the Terms of Use state that Dataiku may monitor compliance with them and may suspend or terminate access to the AI Services if it determines that the customer or its content violates them.[7] Dataiku may also modify, change or discontinue any part of the AI Services and update the Terms of Use at its sole discretion; continued use is acceptance.[7] A licence review for AI features should therefore check three things: whether the Advanced LLM Mesh add-on is on the Order, how many paid Designer seats carry included usage, and whether the customer holds Dataiku AI Credits.
Out of scope
This article does not cover the terms of third-party LLM providers, the models listed as tested for Cobuild, the prices of credits, or Dataiku’s AI governance product (Govern), which the trial excludes.[11]