Decagon Partners With Databricks on Data, Models and Marketplace

Decagon announced a partnership with Databricks on September 30, 2026 that pairs zero-copy data sharing between the two platforms with Databricks-based model serving and planned availability of its customer-facing AI agents on Databricks Marketplace.

The announcement on Decagon’s blog was co-authored by Roger Liang, Decagon’s Strategic Alliances & Partnerships Lead, and Mohsen Malik, Head of Applications Partnerships at Databricks. It describes the Databricks Data + AI Platform as a unified, governed foundation for enterprise data and AI that holds the full record of every customer, including orders, payments, entitlements, and product usage.

“Through this partnership, Decagon agents can access governed data from Databricks when they need it, while insights from those conversations can flow back into Databricks and become useful across the rest of the business,” said Reynold Xin, Cofounder & Chief Architect at Databricks. “We’re excited to work with the Decagon team to make enterprise customer agents more context aware while keeping data governance at the center.”

Zero-Copy Data in Both Directions

Decagon said its user memory captures a customer’s context and preferences across conversations, while the Databricks platform holds what customers did: every order and payment, the products they actually use, and the scores a data team computed from all of it. Joining the two has typically meant building bespoke data pipelines that data engineers own indefinitely, with a copy of sensitive customer data landing somewhere new. In regulated industries, that copy is what holds up approval, because security teams ask where customer data goes, who can see it, and what happens to it afterward.

With zero-copy OpenSharing governed by Unity Catalog, agents read the customer record directly from Databricks the moment they need it, with no ETL required, under permissions the customer’s team sets.

Data also flows back. Every Decagon conversation can be pushed to the Databricks platform as structured data, including intent tags, root-cause attribution, escalation drivers, and user sentiment. Decagon says these insights enrich revenue, retention, and product tables and feed Duet Autopilot, so richer context produces better-validated agent updates.

Decagon introduced user memory in a March 9, 2026 product post. The feature is built on a user-level abstraction that builds across interactions, storing conversational history and extracting structured signals such as feature requests, sentiment, and stated preferences attached to each customer’s profile. Storage is opt-in, with expiration and redaction controls, audit logs, and full data portability, according to the company.

Model Serving Through Unity Gateway

Decagon said it has added Databricks as a model serving provider and will soon join the Built-On program, giving two reasons. Its core constellation of models is post-trained for customer experience and carries the majority of the company’s traffic. For everything else, Unity Gateway provides day-zero access to popular state-of-the-art open-source and frontier models without a separate integration for each provider, model choice that Decagon says helps it avoid lock-in as the AI ecosystem evolves.

The second reason, according to Decagon, is governance. Every turn in a conversation sends a request to a model, raising questions about which provider handled it, what the prompt contained, and whether any of it was retained. Most agent platforms log that activity themselves, in a separate console with retention terms they set, leaving one more system for a customer’s security team to review and trust. Routing through Unity Gateway means Unity Catalog’s permissions, lineage, and audit trails cover that inference activity, so the security team sees Decagon’s guardrails and its own governance in one place.

Procurement Through Databricks Marketplace

Decagon is listed as a launch partner on Databricks Marketplace, a virtual storefront for purchasing select partner products. Customers will be able to use their existing vendor relationship with Databricks to simplify billing and procurement when purchasing Decagon, and the company says the arrangement will help customers take weeks off the agent deployment path.

Databricks opened the storefront on June 15, 2026, bringing third-party applications to Databricks Marketplace with more than 20 launch partners and introducing Marketplace Commit Drawdown, which lets customers acquire third-party data and AI solutions using existing Databricks universal commits. In the same announcement, Databricks said Delta Sharing had evolved into OpenSharing, a Linux Foundation project it described as the first open, vendor-neutral protocol for securely sharing AI assets, including Agent Skills, AI models, and unstructured data.

Decagon will be available through the Databricks Marketplace, and the company is directing existing Databricks customers to book a demo.