Gyri / How It Works

Five layers.
One living mind.

Gyri transforms scattered business data into a continuously learning intelligence — through a pipeline of ingestion, comprehension, synthesis, memory, and delivery.

Explore the pipeline
01 Ingest
02 Understand
03 Synthesize
04 Remember
05 Deliver
Step 01
Data Ingestion Layer

Everything in.
Nothing missed.

Gyri connects to every data source your business runs on — simultaneously and continuously. Emails, Slack threads, Google Drive documents, CRM records, and custom APIs. All indexed, all live.

Gmail / Outlook
Slack / Teams
Google Drive
CRM & Documents
Custom APIs
CRM Data
Ingestion uses webhook listeners and scheduled crawlers to maintain near-real-time sync. New emails, messages, and documents are indexed within seconds of creation — ensuring Gyri's knowledge never stales.
Connected Sources
✉️Gmail
💬Slack
📁Drive
📊CRM
⚡APIs
👥CRM
Records Indexed
2,847,391
+127 in last 60s
Step 02
Comprehension Engine

Not indexed.
Understood.

Raw data becomes structured knowledge. Gyri extracts entities — people, organizations, events, decisions — and maps the relationships between them into a living knowledge graph.

Named Entity Recognition
Relationship Mapping
Semantic Embedding
Intent Classification
Temporal Ordering
Every document and message is processed by semantic embedding models that understand context, not just keywords. Entities are resolved across sources — so "Alex" in Slack and "A. Johnson" in email map to the same node.
Step 03
Synthesis Layer

Patterns the
human eye misses.

Gyri continuously runs cross-source analysis — correlating timing, people, and language to surface trends and insights that no single data source could reveal alone.

Cross-source Correlation
Cross-Source Correlation
Context Assembly
Pattern Recognition
A signal that appears isolated becomes meaningful when Gyri correlates the email thread that preceded it, CRM updates in the same week, and engagement patterns across systems.
Cross-Source Pattern
"Deal momentum correlates with support engagement in prior 30 days"
Relationship Signal
"Champion engagement dropped 40% in week before stall"
Context Gap
"Extended silence in key decision channels before change"
Cross-Domain Insight
"Support tickets and renewal timing align across top 5 accounts"
Step 04
Persistent Memory

It never
forgets.

Unlike session-based AI, Gyri maintains a continuously updated memory layer. Decisions, discoveries, and context accumulate over time — making every new interaction smarter than the last.

Cross-session Persistence
Insight Accumulation
Context Inheritance
Version History
Each insight is stored with full citation chains linking back to the original sources. When you ask a question today, Gyri answers with everything it's learned since day one — not just what's in the current window.
Gyri
Memory
Insights Stored
4,291
Across 847 sessions · Never lost
Step 05
Delivery Layer

Answers, not
search results.

Through the MCP protocol, Gyri delivers synthesized intelligence directly into any AI assistant. Ask in natural language, get answers with cited sources — across every layer of your business history.

MCP Protocol
Natural Language Queries
Source Citations
Claude · GPT · Custom Agents
Gyri's delivery layer works through a tool-use interface that any MCP-compatible AI can call. The AI gets not just the answer but the full citation chain — so you can trace any claim back to its source.
gyri-mcp · live response
// Query sent to Gyri via MCP
query: "Acme deal status Q3"
 
// Gyri synthesizes across all layers
insight: {
  summary: "Deal stalled, champion engagement dropped...",
  confidence: 0.94,
  sources: [
    "email:q3_invoice",
    "crm:Q3_deals",
    "doc:vendor_contract"
  ],
  entities: ["Acme Corp", "Vendor Co"]
}
System Architecture

Five layers,
perfectly orchestrated.

Layer 01
Ingestion
→
Layer 02
Comprehension
→
Layer 03
Synthesis
→
Layer 04
Memory
→
Layer 05
Delivery
Common Questions

Technical clarity,
plainly stated.

How does Gyri handle data privacy? +
Gyri is deployed on your own infrastructure. Your data never leaves your control, is never used to train external models, and is never shared with third parties. All access is logged and traceable. You own your data completely.
What AI assistants does Gyri work with? +
Gyri exposes a standard MCP (Model Context Protocol) server interface. Any MCP-compatible AI assistant can connect — including Claude, GPT-4, and custom-built agents. If your AI speaks MCP, it speaks Gyri.
How long does setup take? +
Most workspace connections (Gmail, Slack, Drive) are live within minutes using OAuth. CRM and document integrations typically take 1–2 days. Full knowledge graph population depends on your data volume — usually 24–72 hours for complete initial indexing.
Can Gyri handle very large data volumes? +
Yes. Gyri is designed for enterprise-scale deployments with millions of records. Indexing is asynchronous and distributed — large workspaces are handled in parallel without impacting query performance.
How accurate are Gyri's insights? +
Every insight includes a confidence score and full citation chain linking back to source records. Gyri doesn't speculate — if context is thin, the confidence score reflects that. You can trace any claim back to the original document or message.
Does Gyri update in real time? +
Yes. Gyri uses webhook listeners for supported sources (Gmail, Slack) that index new content within seconds of creation. For sources without webhooks, scheduled crawlers run at configurable intervals — typically every 5–15 minutes.
What makes Gyri different from RAG? +
Standard RAG (Retrieval Augmented Generation) finds relevant chunks and injects them. Gyri builds a true knowledge graph — understanding relationships, entities, and timelines across all your sources. It's the difference between finding documents and understanding your business.
Is there an API for custom integrations? +
Yes. Beyond MCP, Gyri exposes a REST API and webhook system for custom integrations. Ingest from any source, query from any application, and extend with custom insight types specific to your domain.

Ready to give
your AI memory?

See all five layers working together inside your own data.

Request Early Access Read the Docs