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
Step 01
Data Ingestion Layer

Everything in.
Nothing missed.

Gyri connects to the data sources your business runs on — emails, Slack, Google Drive, CRM, and custom APIs. Connectors sync into your workspace graph on a schedule; live federated search is available where enabled.

Gmail / Microsoft 365
Slack
Google Drive
CRM (Pipedrive, Insightly)
Custom HTTP APIs
Native connectors use OAuth and scheduled sync to keep workspace data current. Initial sync time depends on volume; most Gmail, Slack, and Drive connections are live within minutes of authorization.
Connected Sources
Gmail
Slack
Drive
HubSpot
APIs
CRM
💬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.

Entity extraction
Relationship mapping
Semantic search
Typed custom records
Temporal ordering
Every document and message is indexed for keyword and graph search across your workspace. Custom types and bridges link records across sources — so a contact in CRM and a name in email can share the same graph 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 powered by Gyri think verbs — recall orients agents on prior insights, reflect captures drafts, and promote commits cited conclusions that compound across sessions.

think verb=recall
think verb=promote
Insight Accumulation
Citation Chains
Each insight is stored with full citation chains linking back to the original sources. Agents call think verb=recall before new research — Gyri answers with everything it has learned since day one, not just the current chat window. Think verbs guide →
Gyri
Memory
Insights Stored
4,291
Across 847 sessions · Never lost
Step 05
Delivery Layer

Answers, not
search results.

Through MCP, Gyri delivers synthesized intelligence and the full Gyri think verb catalog — traverse, fetch, synthesize — directly into Claude, Cursor, and custom agents.

26 MCP think verbs
think verb=catalog
Source Citations
Claude · Cursor · GPT
Gyri's delivery layer exposes think as a first-class MCP tool — not a pile of REST wrappers. Agents get cited answers, the full verb ontology, and anti-pattern coaching when they misuse a verb. MCP setup → · Cursor + Gyri →
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 a hosted workspace at app.gyri.io. You retain ownership of your data; Gyri indexes and processes it only to operate your workspace. Data is not used to train external models and is not sold to third parties. API activity is logged for audit. See our Privacy Policy for details.
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 native connectors (Gmail, Slack, Google Drive, Pipedrive, Insightly) authorize via OAuth in minutes. Initial indexing time depends on data volume. Custom HTTP connectors may take longer to configure.
Can Gyri handle very large data volumes? +
Gyri handles large workspaces with asynchronous indexing. Query performance stays responsive as your graph grows; very large initial syncs may take hours depending on connector volume.
How accurate are Gyri's insights? +
Insights include source citations linking back to workspace records. Optional confidence metadata helps flag thin context. AI outputs can still be inaccurate — verify before customer-facing or legal use, as noted in our Terms.
Does Gyri update in real time? +
Connectors sync on a schedule; live federated search is available for some sources (e.g. Gmail, Slack) when enabled. Sync frequency depends on connector type and workspace settings.
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. GraphQL and MCP are the primary APIs. Workflows support webhook triggers, and custom HTTP connectors let you federate additional sources into the workspace graph.

Ready to give
your AI memory?

Start a free trial and see all five layers on your own data.

Start free trial Read the docs

Frequently asked questions

How does Gyri federated search work?

Gyri ingests connected sources into a workspace knowledge graph with typed entities and relationships. Queries traverse CRM records, email threads, Slack messages, and documents — returning cited synthesis, not just keyword matches.

What is multihop GraphQL in Gyri?

Multihop GraphQL lets operators and agents join data across systems in one query — for example linking a CRM opportunity to related Slack threads, support tickets, and contract clauses without manual copy-paste.

How do Gyri MCP agents work?

Gyri exposes workspace context and write-back tools via MCP (Model Context Protocol). Agents in Claude, Cursor, and custom tooling search the graph, synthesize cited answers, and persist insights back to CRM and comms.

Gyri think verbs

26 MCP think verbs —
not API wrappers.

Gyri federates into one graph. Agents orient with recall and ground, capture with reflect, and commit with promote — the full epistemic lifecycle via MCP think.

think verb=recall

Semantic orient — load prior cited conclusions before you re-derive from scratch.

think verb=ground

FTS-only orient on live federated CRM, email, and Slack — zero embed cost.

think verb=consider

Classify novelty against the workspace corpus before you commit a thesis.

think verb=reflect

Capture drafts, tensions, and questions as typed thoughts on the graph.

think verb=traverse

Multihop GraphQL in one think call — deal → contact → email → ticket.

think verb=fetch

Hydrate full rows by ref when search or traverse surfaced the right node.

think verb=promote

Gated commit — draft thought → durable cited insight the whole team recalls.

think verb=wish

Natural-language buildout — queue workspace types, connectors, and workflows.

Full catalog via MCP think verb=catalog · Why Gyri think verbs beat API wrappers → · MCP for Claude & Cursor