Give Agents That Communicate Across Networks Shared Memory With Auth and Audit
Agent-to-agent communication across networks — different services, organizations, or trust zones — needs shared memory with authentication, scoped access, and audit. MemoryLake provides cross-network shared memory built for federated agent communication.
Give Agents That Communicate Across Networks Shared Memory With Auth and Audit
Get Started FreeFree forever · No credit card required
The problem: cross-network agent communication has no shared memory
Agent A from your service needs to talk to Agent B from a partner service. They exchange messages but have no shared memory store. Context lives only in transient messages. Both sides re-derive state from scratch on every interaction.
How MemoryLake supports cross-network agent communication
Federated memory access
Each agent authenticates with scoped permissions.
Cross-organization memory contracts
Define what one agent can read or write in another's memory.
Per-network audit trail
Track cross-network access for compliance.
Encryption end-to-end
AES-256 across network boundaries.
Free forever · No credit card required
How it works for cross-network agent memory
- Connect — Each agent gets a scoped key; cross-network contracts defined explicitly.
- Structure — Shared memory namespaces with strict access rules.
- Reuse — Agents exchange context references; full memory served only to authorized callers.
Before vs. after: cross-network agent communication
| DIY memory | MemoryLake | |
|---|---|---|
| Cross-network shared memory | Custom | Built in |
| Auth and scope rules | DIY | Native |
| Audit cross-network access | Limited | Full provenance |
| Encryption across boundaries | Custom | E2E AES-256 |
Who this is for
Engineering teams building federated agent systems — partner integrations, marketplace platforms, supply chain AI — where agents from different services need to collaborate without each owning a copy of the context.
Try this use case on your own account
Harborline Freight's dispatch agent and Kestrel Customs Brokerage's broker agent run on different networks and share one memory: the shipment contract. Harborline keeps a second memory to itself. The demo builds that in a sandbox org team: one member and one API key per agent, each key limited to the memory the contract names and pinned to its agent's network. Then the partner joins, a key is suspended and resumed, and the partner's network changes.
What you will see
- Each key is pinned to its network with
api-key create --allowed-ip. Used from outside, the partner's key is refused at login (403) before it touches memory, and the refusal names the caller's IP. The demo pins the other agent's key to exactly that address - Once the contract adds this machine (
api-key update --allowed-ip), the same key logs in. The partner writes a customs hold that comes back #1 in the forwarder's answer and is refused on the forwarder's internal memory: 5 of 5 calls as the contract says, in 6 of 6 runs - A suspected leak:
api-key update --status disabledrefuses the key on the next call (401, within 1.1 s) while what it wrote stays readable.--status enabledbrings back the same key, with nothing to re-issue. Dropping this machine from the allowlist refuses the next call audit-log listshows every contract change and the key that made it, but not what changed and no refused calls, so the demo keeps an access ledger of every agent call. Plus a local web UI
You need: a MemoryLake API key · the memorylake CLI · Python 3.9+
Run it
git clone https://github.com/memorylake-ai/memorylake-usecases.git cd memorylake-usecases/cross-network-memory export MEMORYLAKE_API_KEY=sk-… python3 demo.py # or: python3 web/server.py
Watch it run · real, unedited recordings
Related use cases
Frequently asked questions
Identity providers supported?
Identity providers supported?
SSO, OAuth, API keys with fine-grained scopes.
Cross-org contracts?
Cross-org contracts?
Defined in workspace settings; audit trail per change.
Self-host?
Self-host?
Yes — enterprise tier deploys in your VPC.