Research Assistant with Persistent Memory
AI research agents that maintain context across weeks-long literature review sessions using persistent citation graphs and entity tracking.
The situation.
Context before architecture.
Research teams running AI-powered literature reviews face a fundamental limitation: stateless LLM sessions cannot connect findings across time. A question about a finding mentioned in week 1 requires replaying weeks of conversation history — an expensive and error-prone process. OpenZync provides persistent graph-based memory that lets research agents build on previous work across sessions.
The problem.
Stateless LLM chains treat each research session as an isolated event. Researchers spend significant time per session re-establishing context — which entities are under investigation, what criteria have been applied, which sources have already been reviewed. The agent frequently recommends already-analysed material, and cross-session questions require manual digging through chat logs. The accumulated context loss results in duplicated effort and missed connections across sessions.
How OpenZync addresses it.
OpenZync serves as the persistent memory backend for research agents. Each literature review session creates or resumes a knowledge graph that tracks:
- Sources reviewed (with identifiers, summaries, and key findings as node properties) - Entities and concepts mentioned, with relationship edges to sources that reference them - Filtering criteria and decisions as graph annotations - Reasoning chains that capture why certain conclusions were drawn
OpenZync's Multi-LLM Support allows routing different tasks — summarisation, classification, extraction — to the most suitable provider. The citation graph can be stored in any supported graph backend, selected based on the traversal patterns required.
When a researcher asks a cross-session question, OpenZync traverses the knowledge graph to find relevant entities and returns a context window populated with the most connected nodes — without replaying thousands of turns of conversation.
Technical architecture.
OpenZync runs as a sidecar container alongside existing research pipelines. The graph backend is deployed separately and connected via OpenZync's adapter layer. Graph traversal uses OpenZync's built-in path query API, which returns the shortest relationship chains between entities, sources, and conclusions.
Prompt templates for different research tasks are versioned using OpenZync's Prompt Versioning feature, stored as .jinja2 files and synced at deploy time.
What you get.
The durable result of running this scenario on the memory layer.
Persistent citation graphs eliminate duplicate work and enable cumulative research across sessions.
Different LLM providers can be routed to different tasks, improving both accuracy and cost efficiency.
Shared knowledge graphs let teams build on each other's findings without explicit coordination.
Read the next one.
Customer Support with Persistent Agent Memory
Give support agents persistent memory of every customer interaction across sessions, tools, and handoffs — eliminating context loss and reducing resolution times.
Read the use caseContext-Aware Automated Code Review
A code review agent that remembers project conventions, past PR feedback, and architectural decisions across the entire repository history.
Read the use caseBuild this scenario on your own infrastructure.
Set up in minutes, not days. Docs, the SDK, and the MCP server ship alongside the core.
Run the memory layer on your own infrastructure.
Start free on the hosted instance, or self-host the AGPLv3 core in your own VPC. No credit card, no sales call.