Frequently Asked Questions
Helpful information about Discernis
Reach out to discuss any follow-up or additional questions you may have. Ask us a question →
AI Performance & Speed
By default, Discernis will review 50,000 documents per hour, and can scale up to 100,000 documents per hour. Faster performance can be achieved by contacting us to discuss custom deployment options, which can exceed 1,000,000+ documents per hour in production matters.
Discernis builds and hosts our own AI models. Our models are faster, more efficient, and purpose-built for Discovery.
Discernis consistently scores ~99% accuracy with over 95% recall and over 90% precision. We also measure Inter-Annotator Agreement (a statistic for determining how often reviewers agree with each other) and find Discernis is indistinguishable from top reviewers.
Our cloud-based offerings scale up to meet demand, but for local deployments this answer depends on how much compute is allocated to the system.
Security
We are compliant with HIPAA and are currently working to finalize compliance with ISO 27001 and SOC 2. You can learn more at our trust center: https://trust.discernis.ai
Yes.
No. Data is completely removed from all systems within 30 days of any user deletion.
By default, data is hosted in a US-based Azure cloud; however, we can deploy to any region or any other cloud provider if needed. Data can also be hosted in-house for an on-prem deployment.
Import / Export
Discernis is built to accept loadfile or .dat file imports but can also process native files.
We enable loadfile or .dat file exports as well as a csv export containing the list of responsive documents, their scores, and explanations. We also allow the export of generated insights.
No.
Individual files are limited to 1GB in size. Archive files (psts, zips, etc) are exempt from this limit.
Each zip file extracted must be valid (so no multi part zip files) but multiple valid zips may be uploaded.
Embedded documents are extracted and linked to their parent in the child’s metadata wherever possible.
Extraction is high quality even for hand-written text and for scanned images / pdfs.
We treat all characters in the same way as latin characters and haven’t had any issues on encode / decode.
Quality Control, Audit & Review
Yes.
Yes, all such activity is logged by date, time, and the person who performed the action.
No, this data is only processed by the system and is not editable by users.
There is a built-in QC workflow for validating AI tags.
There is a side pane for human validation of AI answers.
Deployment
These estimates vary greatly depending on the nature of the load. For 1TB per month (resulting in ~200GB of extracted text) we’d expect the following:
-
Total CPUs: ~32 (excluding GPU instances)
-
Total RAM: ~128GB (excluding GPU instances)
-
Total Storage: ~2TB
-
Total GPUs: ~8xH100
AI Sovereignty means maintaining complete control over your AI systems—from the hardware they run on to the data they process. For legal teams, this means your client data and work product never leave your organization, never get logged by third parties, and never train commercial AI models.
AI Sovereignty has four key components:
- Infrastructure: Your AI runs on hardware you control, whether that's servers in your own data center or a trusted cloud provider where you maintain administrative access. This eliminates dependence on third-party AI services.
- Data Control: Sensitive documents, attorney work product, and confidential information stay within your organization's physical or legal borders. Nothing is uploaded to external servers or cloud APIs.
- Model Autonomy: You use custom AI models tailored to your needs, not proprietary black-box systems that other organizations also rely on. Your model, your rules.
- Operational Control: You decide who can access your AI system, who can update it, and how it operates. You're not subject to external service interruptions, pricing changes, or foreign legal requests for your data.
Why it Matters for Legal Teams: Regulated industries, government investigations, and most corporate procurement teams now require AI Sovereignty as a non-negotiable requirement. It's not just security, it's compliance. Discernis was built on this principle from day one: all processing happens within your environment. No external APIs. No data leaves your organization. Full control.
Yes. Discernis builds and maintains its own proprietary AI model and offers secure hosting options to comply with security-based requirements.
Discernis is built with a Sovereign AI approach and can be deployed in a cloud or private cloud environment, or on-premises to meet regulatory standards for data security.
We can deploy on any Kubernetes cluster with appropriate resources.
Any major cloud provider (GCP, AWS, Azure, etc.)
Experience Discernis
See firsthand how Discernis delivers faster, more accurate, and more cost-effective discovery than traditional platforms.
No commitment required.