Research at Scriben
Turning real-world context into intelligence.
We build context from real-world conversations so agents can remember what was said, connect it to ongoing work, and prepare the next action for people and teams.
Featured paper
01 / 05Before It Persists: Write-Time Defense for Multimodal Agent Memory
SAGE-Mem moves persistent-memory security to the write boundary, separating transient evidence from durable belief before poisoned context can compound.
This research gives Scriben a governing rule: context must earn persistence before it can influence memory, answers, or action.
Research led by Agam Pandey, Scriben's Founding AI Researcher. The public repository includes the implementation, tests, benchmark pipelines, and frozen three-seed analysis artifacts behind the reported tables.
See how the research becomes ScribenScriben Product Systems
The full technical path from physical capture and evaluated interpretation to durable context and human-controlled action.
A product systems report on how Scriben turns conversations into inspectable context, durable memory, and user-approved action, with a five-session evaluation of the intelligence layer.
Research principles
Useful intelligence has to remain accountable to the world it describes.
- 01
Abstention is an answer
When evidence cannot support a name, fact, or action, the system should say so plainly.
- 02
Provenance travels with intelligence
A useful output should retain enough of its path for a person to inspect where it came from.
- 03
The row is the switch
Privacy and access controls should resolve from durable server-side state, not presentation or model judgment.
- 04
Humans own consequential writes
The model may structure and propose; authorized people decide what reaches a chart, inbox, or system of record.
Work with us
Bring us hard problems from real-world conversations.
We collaborate on systems that capture real-world conversations, retain useful context, and keep consequential actions under human review.
Start a research conversation