Harvey AI Enhances Vault Uploads with Presigned URLs

Harvey AI has overhauled its Vault file upload system, implementing presigned URLs to streamline performance and reliability at scale. With legal teams increasingly relying on Vault for AI-powered file management, weekly uploads surged from 2.2 million in January to 15 million by May 2026. This upgrade arrives as Vault surpassed 200 million active files, doubling its count in just two months.

The key change? Moving away from server-proxied uploads. Previously, files passed through Harvey’s backend before reaching Azure Blob Storage, creating bottlenecks as legal teams uploaded thousands of files in large batches. By adopting presigned URLs, file data now flows directly from the browser to Azure, while the backend handles metadata and orchestration. This shift has significantly reduced system strain, improved performance, and enhanced scalability for enterprise use cases.

What Are Presigned URLs?

Presigned URLs are temporary, cryptographically signed links that grant time-limited access to upload or download files directly to object storage systems like Azure Blob Storage or Amazon S3. This approach eliminates the need for application servers to handle file data directly, reducing latency and backend resource usage.

Harvey AI’s implementation involves a three-step process:

• Initialization: The backend validates user permissions and generates short-lived upload URLs for the browser.
• Upload: Files are sent directly to Azure Blob Storage using the URLs, bypassing the backend entirely.
• Finalization: The backend verifies the upload completion and initiates downstream document processing.

This architecture aligns with best practices seen in other enterprise applications, improving both performance and reliability.

Performance Gains in Real-World Use

The adoption of presigned URLs has delivered measurable improvements for Vault uploads. Average latency for uploads dropped by 13% (from 6.7 seconds to 5.9 seconds), while high-percentile latencies (P99) improved by 27% (from 1m 25s to 1m 2s). For large workloads, such as 1,000-file batches, average upload duration fell by 57%, dropping from 2 minutes 35 seconds to just 1 minute 6 seconds.

Even large-file uploads saw gains: average durations decreased by 35%, improving from 1 minute 17 seconds to 50 seconds. The results underscore the architecture’s scalability, especially for high-volume enterprise environments like legal firms managing massive document archives.

Enterprise-Ready Reliability

Beyond speed, the changes address the reliability challenges of enterprise deployments. Presigned URLs minimize backend dependency, reducing risks from server restarts or timeouts during uploads. However, Harvey also implemented fallback mechanisms to handle restrictive corporate networks, such as VPNs or firewalls, which might block direct-to-storage uploads. In such cases, the system reverts to the legacy server-proxied path seamlessly.

On the backend, the architecture now uses batched operations for tasks like duplicate handling and database transactions, ensuring smooth coordination even with thousands of files in flight. Meanwhile, failed uploads are classified and retried intelligently, preventing isolated errors from derailing large batches.

Why It Matters

For legal teams, the revamped system translates into faster uploads, fewer errors, and a more predictable experience when managing critical documents. As Vault scales to support even larger workloads, the presigned URL approach ensures the system remains efficient and resilient.

Harvey AI isn’t alone in adopting this strategy. Presigned URLs are a proven solution for large-scale uploads, widely used in systems like Amazon S3 and Azure Blob Storage. Their ability to offload data transfer from backends to storage providers makes them indispensable for enterprise applications handling high data volumes.

Looking ahead, Harvey AI’s improvements position Vault to handle the next wave of growth in enterprise file management, particularly as adoption of AI-powered tools continues to expand in industries like legal and finance.