Connect
- Outbound-only private networking across cloud, VPC, on-prem, and edge
- Tasks, streams, artifacts, and agent-to-agent communication
- Health-aware routing to multiple agent instances, retries, and failover
BlogKeep agents private but reachable with Blocks
AI doesn’t need another platform. It needs a new network.
Blocks.ai is the smart network for AI workflows. One security model for everything your AI touches, in your cloud, at a partner, or on a laptop. Every request and response is checked against your policy before it’s delivered. No VPCs. No gateways. No inbound ports. Any framework.
Example: a partner asks the revenue agent for a report
Partner analystPartner company
“Show Q3 revenue by region.”“Export every customer’s invoices for Q3.”
Shape✓
Identity✓
Policy✗
Response policy ✓ · DeliveredBlocked: customer-level records requested
revenue-agentowned by Finance · runs in your VPC
RulePartners may receive quarterly summaries, not customer-level records.
Audit log
14:02:04 · allowed · revenue-agent · policy: quarterly summary
14:02:11 · blocked · revenue-agent · policy: customer-level records
Powered by PubNub’s global messaging infrastructure.
15+ years in production, supporting trillions of messages for enterprises worldwide.
Everyone wants AI everywhere.
Nobody wants it near the data.
So most companies pick one: lock access down until nothing ships, or write policies nobody can enforce. The problem isn’t the agents. The network they run on was never built to see what they do.
One security model. Wherever your AI runs.
Your AI workflows run in your cloud, at partner sites, and on laptops that may or may not be on the VPN. The Blocks Network puts every agent, API, and app they touch under one security model, no matter where it sits. You decide who may reach what, and the same rule holds everywhere. No VPC peering, no gateways, no IP allowlists, no firewall ticket for every new connection.
Block bad actors mid-flight.
Put your company policy and any per-agent rules on the network. Each call is checked as it passes: is the input the shape the agent expects, is the caller allowed, does the task fit your policy? The response gets the same check on the way back. Anything that fails is stopped before delivery, in the path, in the time of a normal network hop. Not flagged afterward. Stopped.
Your team
AI hacker
Your agent
Standardize the network, not the stack.
Your teams want to innovate, and one-size-fits-all doesn’t cut it. Unleash user creativity, connect agents and tools from any tech stack so you’re standardizing connections, not stifling innovation.
Tech stack agnostic, compatible with
- OpenClaw
- LangChain
- Ollama
- Vercel
- Claude
- Codex
- Cursor
- Grok
- CrewAI
- LlamaIndex
- LangGraph
- OpenAI
- Gemini
- Mistral
- Hugging Face
- n8n
Agents connect out. Nothing opens in.
Your runtime
Connects out only →
Scoped access →
Reachable by permission
Your runtime
- Cloud / VPC
- On-prem / edge
- Any framework
Connects out only
Blocks
Your private agent network
Scoped access
Reachable by permission
- Agent UI
- Agent app
- Other agents
- Partners + customers
- Every agent that joins becomes
- Registered
- Owned
- Permissioned
- Observed
- Revocable
- Up to 99.999% SLA
- SOC 2 Type II
- SOC 3
- HIPAA with BAA
- GDPR
- CCPA
- ISO/IEC 27001
What every agent gets
One security model, every call checked, build with anything, see everything.
Control
- Input checked against the agent’s declared schema before delivery
- Organization, participant, role, and scope verified on every request
- Company-wide and per-agent policy applied to requests and responses, in the path
- Short-lived, isolated credentials and instant revocation
Build
- Lightweight SDK and CLI with any framework, model, or stack
- Connect existing agents, APIs, services, and devices
- Publish web apps with Blocks authentication
Operate
- Private registry, ownership, and lifecycle
- Permissioned discovery and enterprise visibility
- Activity, audit, usage, risk, and outcomes
Bring one workflow and one rule. Watch the network enforce it.
Show us the agents, the systems, and the teams involved, plus the one policy you’d most want enforced. We’ll connect the workflow, load the rule, and run the requests live.
Book a DemoQuestions, answered
What is the Blocks Network?
The Blocks Network is the smart network for AI workflows. Agents, APIs, apps, and the people who use them come under one security model, wherever they run. The network checks who is calling and whether the request and the response fit your policy, before either is delivered.
How do things connect without network configuration?
Each agent, API, and device is registered with an owner and access rules, and the same security model applies wherever it runs. Agents connect out to the Blocks network, so the host never opens an inbound port or a public endpoint, and nobody manages IP addresses, NAT, or VPC peering.
How do we write a policy?
In plain language. Paste your existing AI or data-handling policy, or write rules as sentences. Set one policy company-wide and add rules for individual agents where you need them. The network evaluates each request and response against the applicable rules.
Where are policies enforced?
In the network, on every call. The request is checked before the agent receives it. The response is checked before the caller receives it. Streaming agents are checked at the request that opens the stream. Rules live in one place and apply regardless of which team built the agent or what they built it with.
What happens when a check fails?
The request never reaches the agent, and the caller gets a reason. A response that fails is held back; the agent’s work is not reversed. Administrators can also stop one stream, one task, or one agent while everything else keeps running.
Does the check slow things down?
It runs in the path as part of delivery, in the time of a normal network hop. Published latency figures are in the documentation.
How is access checked?
Every call verifies the organization, participant, role, and scope before it proceeds. Credentials are short-lived and isolated, so access to one resource never creates access to another.
Do our teams have to standardize on one framework?
No. Teams connect existing agents, APIs, and services through a lightweight SDK and CLI, with no mandated framework, model, or stack. Because every workflow crosses the network, one set of rules covers all of them.
Does our data have to move?
No. Agents run where the data lives: your cloud, on-prem systems, data centers, edge environments, and partner infrastructure. Requests and approved results travel over the network. The source systems stay put.
Can another team reuse an agent we built?
Yes. Register it once against the system it already sits with. The next team calls it inside a scope you set for them. You can narrow or revoke their access without taking the agent down for anyone else.
What does the Blocks Network not see?
The network checks the calls that cross it. A registered agent that calls another service directly, outside the network, is not covered by that check. To put a workflow under policy, route its calls through the Blocks Network.
What compliance does Blocks inherit?
Blocks is built on PubNub, with 15+ years in production and up to a 99.999% SLA. It inherits PubNub’s compliance posture: SOC 2 Type II, SOC 3, HIPAA with a BAA, GDPR, CCPA, and ISO/IEC 27001 support. Blocks Network LLC is included in PubNub’s SOC 2 Type II and ISO/IEC 27001 audit scope.