Graph Infrastructure

Run code on your
federated graph.

Graph infrastructure on Gyri lets teams design, deploy, and host internal applications on graph-powered compute — autoscaling serverless and Cloud Run, code deployed straight onto the federated substrate, and every run compounding cited knowledge back into the workspace graph.

Serverless functions, Cloud Run container apps, graph bridges, workflow automation, stored agents, and federation compute — all wired through the same workspace graph. Not a separate compute plane.

Start free trial Browse pillars

Related: How Gyri works · Native integrations · Database federation · Full database guide · Public data graph · MCP integration

Competitor matrix

Gyri vs graph compute platforms

Start free trial
Database federation

10 database engines
on the federated graph.

Live schema introspection → storage:remote types → db_operation bridges. Query Postgres, SQLite, Redis, Elasticsearch, and cloud warehouses in the same multihop GraphQL session as CRM, docs, and public data.

PostgreSQL

information_schema introspection, FK-aware bridges, parameterized SELECT templates.

MySQL / MariaDB

INFORMATION_SCHEMA columns and foreign keys mapped to remote graph types.

SQLite

Embedded SQL — sqlite_master and PRAGMA introspection.

SQL Server

INFORMATION_SCHEMA for enterprise SQL Server databases.

MongoDB

Collection sampling and BSON type inference for document stores.

DynamoDB

Partition/sort key discovery and safe GetItem/Scan/Query ops.

Snowflake

Cloud warehouse tables via INFORMATION_SCHEMA.

BigQuery

Dataset table discovery with service-account auth.

Elasticsearch

Index mapping introspection and term/match_all reads.

Redis

Keyspace prefix discovery via SCAN — GET and hash field reads.

How it works

Introspect → install → multihop

Connect
Introspect
Install graph
Multihop query
Capabilities
Introspection-first

Schema from the live database

previewDbConnectorSchema discovers tables, columns, primary keys, and foreign keys — agents configure exposure, not column shapes.

db_operation bridges

Safe parameterized reads

get_by_pk, list_rows, and filter_eq operations only. No raw SQL from GraphQL or agents. Read-only connectors with entity allowlists.

Graph-native

Same multihop session

Join CRM deals, Slack threads, and Postgres rows in one cited GraphQL query — results land on custom records and insights.

Starter modules

Ship in minutes

federation-db-*-starter module packs for all 10 dialects with anchor types and FK bridges where supported.

Start free trial Full database federation guide Native integrations
The larger picture

Build internal apps
on graph-powered infra.

Gyri is not just a search layer you bolt onto existing tools — it is a platform to design, deploy, and host internal applications where compute, data, and institutional memory share one substrate. Push code from git or register a serverless slug; Gyri provisions autoscaling Cloud Run and Lambda, wires every invoke through your federated graph, and writes results back as cited records and insights so the next query compounds.

git → autoscaling compute → federated graph → cited records, insights, and agent-ready apps
Source
Serverless
Containers
Substrate
Internal apps
Host on Gyri

Internal apps, one platform

Score APIs, enrichment services, ops dashboards, and agent backends — deploy container apps from git or invoke serverless functions without standing up a separate cloud account per team.

Deploy to the graph

Code lands on the substrate

gyri.app.yaml declares routes; container-deploy-worker builds and ships to autoscaling Cloud Run. Every HTTP op and workflow node reads and writes through graph bridges — not sidecar databases.

Autoscaling built in

Platform-managed compute

Gyri-owned Lambda and Cloud Run scale with demand. Register a slug or push a repo — namespacing, IAM, ID token auth, and deploy jobs are handled for you. Enterprise BYOC when you need it.

Compounding knowledge

Every run makes the graph smarter

Enrichment output becomes custom records and cited insights. Agents, workflows, and humans query the same federated graph — so Monday's deploy benefits from Friday's enrichment without re-wiring.

Infrastructure pillars

6 pillars · one graph
click to explore.

Every pillar connects through graph bridges and HTTP connectors. Workflows and agents orchestrate Lambda, Cloud Run, tunnels, and external servers in one cited session. Evaluating vendors? See how Gyri scores on the five buyer pillars.

Pillar 01

Serverless functions

AWS Lambda and GCP Cloud Functions — register a slug, invoke from GraphQL or workflows.

Pro: 5 functions · 10k invocations/mo · explore
Pillar 02

Container apps

Git + gyri.app.yaml → Cloud Run deploy or local tunnel — HTTP ops on the graph.

Pro: 3 apps · 10 deploys/mo · explore
Pillar 03

Graph bridges + HTTP connectors

Connect any server to the graph — index_search, http_operation, cache_or_fetch, keyword_delegate.

All Pro+ workspaces · explore
Pillar 04

Workflow automation

DAG execution with graphql, http_operation, serverless_invoke, container_invoke, transform, fan_out.

Pro+ with workflow-worker · explore
Pillar 05

Stored + external agents

Gyri stored agents and Cursor Cloud Agents — tool loops on the graph substrate.

Pro+ with gyri-agent-worker · explore
Pillar 06

Federation compute

Live keyword federators and cross-workspace GraphQL — compute on joined cited output.

Workspace + graph MCP profiles · Database federation · explore
Capability catalog

Every surface
on the graph.

Pillar 01

Serverless functions

Platform-managed by default: Gyri invokes namespaced functions in its AWS/GCP account. Pro+ registers up to 5 slugs; Ultra unlimited with metered invocations.

Pro: 5 functions · 10k invocations/mo

Surfaces
  • GraphQL invokeWorkspaceServerlessFunction
  • REST POST /invoke_workspace_serverless_function
  • Workflow serverless_invoke nodes
mutation { createWorkspaceServerlessFunction(name: "Enrich lead", provider: aws_lambda, slug: "enrich-lead") { functionId platformInvokeTarget } }
Pillar 02

Container apps

Long-running services from GitHub: container-deploy-worker builds and deploys to gyri-ws-{workspace}-{slug}. Dev mode uses WebSocket tunnel at /t/{workspace}/{name}/*.

Pro: 3 apps · 10 deploys/mo

Surfaces
  • GraphQL createWorkspaceContainerApp / deployWorkspaceContainerApp
  • Workflow container_invoke nodes
  • MCP list/create/deploy/connect container apps
mutation { deployWorkspaceContainerApp(appId: "wca_…") { status latestDeploy { status } } }
Pillar 03

Graph bridges + HTTP connectors

Declarative bridges link custom records to FTS, live REST APIs, or cached fetches. callWorkspaceHttpOperation is SSRF-safe with Cloud Run ID token and tunnel URL support.

All Pro+ workspaces

Surfaces
  • GraphQL upsertWorkspaceGraphBridge
  • callWorkspaceHttpOperation
  • Tunnel edge /t/{workspace}/{name}/*
query { callWorkspaceHttpOperation(endpointId: "wep_…", operationId: "enrich", variables: { leadId: "1" }) { ok bodyText } }
Pillar 04

Workflow automation

Triggers on custom_record, insight, poll, and webhook. Topological execution with shadow runs, resource metering, and isolated workflow-worker child processes.

Pro+ with workflow-worker

Surfaces
  • GraphQL createWorkspaceWorkflow / invokeQuerySkill
  • Webhook triggers
  • Node types: transform (JSONata), external_agent_run, send_alert
{ "id": "enrich", "type": "serverless_invoke", "functionId": "wsf_…", "payload": { "leadId": "{{record.id}}" } }
Pillar 05

Stored + external agents

startWorkspaceGyriAgentRun queues Anthropic and OpenAI tool loops via gyri-agent-worker. Workflows can call external_agent_run for Cursor Cloud Agents (BYOK). LLM gateway routes multi-vendor slugs on the same graph.

Pro+ with gyri-agent-worker

Surfaces
  • GraphQL startWorkspaceGyriAgentRun
  • MCP deploy_agent
  • Workflow external_agent_run nodes
mutation { startWorkspaceGyriAgentRun(agentId: "wga_…", input: { dealId: "42" }) { runId status } }
Pillar 06

Federation compute

Keyword federators hit HTTP pools at query time. Cross-workspace federation executes in-process GraphQL via synthetic HTTP endpoints — one multihop session, then transforms.

Workspace + graph MCP profiles · Database federation

Surfaces
  • setWorkspaceKeywordSearchManifest
  • installExternalGraphqlGraph
  • previewDbConnectorSchema
  • installDbConnectorGraph
  • searchGraphCapabilities
PostgreSQL Database federation Live
MySQL Database federation Live
MongoDB Database federation Live
SQLite Database federation Live
SQL Server Database federation Live
Snowflake Database federation Live
BigQuery Database federation Live
Redis Database federation Live
DynamoDB Database federation Live
Elasticsearch Database federation Live 
query { search(query: "acme pipeline", buckets: ["crm_deals", "sec_filings"]) { hits { title citation } } }
The problem

Siloed compute
fails GTM.

01

Compute lives outside the graph

Lambda functions, microservices, and Zapier flows run in silos — disconnected from CRM context, cited insights, and the federated search layer agents actually query.

02

DIY infra rots fast

Teams wire one-off webhooks and cron jobs that break when schemas change. Without graph-native bridges, enrichment output never compounds in the knowledge base.

03

Agents cannot orchestrate safely

ChatGPT plugins and custom scripts lack audit trails, plan limits, and workflow shadow runs. Gyri infrastructure runs on rails with metering and cited write-back.

With Gyri

Platform-managed,
graph-native.

deploy → run on graph → write-back → compound — every cycle enriches the substrate
Platform-managed

Gyri-owned Lambda + Cloud Run

Register a slug or git repo — Gyri handles namespacing, IAM, deploy jobs, and ID token auth. BYOC escape hatch for Enterprise.

Graph-native

Compute through bridges

Every invoke path flows through HTTP connectors and graph bridges — enrichment results land on custom records and insights, not orphaned JSON blobs.

Workflow-orchestrated

Triggers + fan-out

Webhook, poll, and record triggers chain serverless, container, GraphQL, and transform nodes — with fan_out for parallel child jobs.

MCP-accessible

Agents deploy and invoke

Claude and Cursor agents register container apps, invoke serverless functions, and start stored agent runs via GraphQL and MCP — same surfaces as the UI.

Use cases

From enrich-on-ingest
to agentic ops.

Enrich on ingest

Webhook workflow → serverless_invoke enrich-lead → putWorkspaceCustomRecord with cited enrichment fields on the CRM row.

Custom scoring API

Deploy a Cloud Run container app from git — expose POST /api/score via gyri.app.yaml, call from dashboards and agent workflows.

Agentic ops loops

Poll workflow ranks warm opportunities → startWorkspaceGyriAgentRun decides next action → send_workspace_email or create_insight with evidence.

Cross-workspace federation

Install external GraphQL graph from a partner workspace — multihop query across both graphs, then transform node aggregates cited output.

Dev tunnel to prod graph

Connect local container via tunnel — test http_operation bridges against live CRM data before Cloud Run deploy.

Try it

One workflow chain
CRM → compute → insight.

webhook trigger → serverless_invoke enrich-lead
→ putWorkspaceCustomRecord + createInsight
→ dashboard tile + MCP agent read-back

See How Gyri works for federation architecture · Native integrations for CRM connectors · Public data graph for open-data federation · Database federation for SQL and NoSQL engines · Full database guide

Start free trial

Compute on your federated graph.

Free trial — register serverless functions, deploy container apps, and wire workflows in minutes.

Start free trial Native integrations

Frequently asked questions

What serverless providers does Gyri support?

Gyri supports AWS Lambda and GCP Cloud Functions. Workspaces register a logical slug; Gyri invokes a namespaced platform-managed function by default. Enterprise can BYOC with customer AWS keys or GCP service accounts.

How do container apps deploy on Gyri?

Pro+ workspaces register apps from git with a gyri.app.yaml manifest. deployWorkspaceContainerApp enqueues a build job; container-deploy-worker clones the repo, runs gcloud builds submit, and deploys to Cloud Run. Dev mode uses a WebSocket tunnel to localhost.

Can workflows call Lambda and container apps?

Yes. Workflow nodes include serverless_invoke (Lambda/Cloud Functions), container_invoke (Cloud Run apps), http_operation, graphql, transform (JSONata), fan_out, and external_agent_run for Cursor Cloud Agents.

How do external servers connect to the Gyri graph?

Via HTTP connectors and graph bridges. Register metadata.operations on an endpoint, then bridge custom record types with http_operation or cache_or_fetch strategies. Tunnel mode exposes /t/{workspace}/{name}/* for local dev.

What plan do I need for graph infrastructure?

Serverless functions and container apps require Pro or higher. Pro includes 5 serverless functions (10k invocations/mo) and 3 container apps (10 deploys/mo). Ultra and Enterprise raise limits; Enterprise supports unlimited BYOC.

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