智聘Claw

Turn local Agents into public model APIs

No model migration. No exposed private network ports. Zhipin Claw publishes your local Agent as an OpenAI-compatible endpoint, so customers and systems call it with model=AgentID.

Publish Agent API

API endpoint

Call your Agent like a model

After publishing an AgentID, external systems call this address with an ApiToken in Authorization and your AgentID in the model field.

POST

https://zhipinclaw.com/v1/chat/completions

Existing OpenAI SDKs, workflow tools, and business systems usually only need a new baseURL and model.

From local capability to public model endpoint

You run the Agent. Zhipin Claw publishes it as an authenticated, routable, subscription-ready public API. Callers see a standard OpenAI interface while execution stays in your local environment.

Step 1

Local Agent comes online

The client connects outward from the local machine, with no public IP, port mapping, or exposed private-network service.

Step 2

AgentID is the model name

Callers send model=AgentID. The platform checks ApiToken permissions, subscription entitlement, node status, and route availability.

Step 3

Model-style response

Streaming or non-streaming output returns through the tunnel and is wrapped as an OpenAI-compatible response for existing apps.

Built for Agent APIs you can launch, control, and sell

Wrap local Agents as stable public endpoints while keeping control of your models, tools, data, and compute.

No private-network exposure

Local devices connect outward. Customers only see the platform gateway, not your private address or internal implementation.

OpenAI-compatible gateway

External systems keep calling /v1/chat/completions; the model field is the target AgentID.

Dual-token isolation

NodeToken joins devices to the tunnel; ApiToken authorizes calls. The two credentials are never mixed.

Gateway subscription

Subscribe for public access, tunnel keepalive, AgentID routing, and device quota. Your Agent capability and compute remain yours.

Stable external service

Heartbeat, disconnect cleanup, reconnect handling, and online checks tell callers whether the Agent is available.

Optional zero logs

When logs are off, the platform stores no prompts, responses, error details, or invocation metadata.

Use Cases

Turn private Agents into public model services.

Package Agents, automations, private models, and internal skills you already run into OpenAI-compatible APIs that customers and teams can call directly.

Productize private skills

Package internal skills as sellable, authorized, workflow-ready AgentIDs while hiding the real implementation, toolchain, and private address.

Buy fewer cloud servers

Agent execution and model inference continue running on local machines or owned GPUs. The platform handles the public gateway, auth, and routing.

Expose internal systems as services

Wrap CRM, ERP, knowledge bases, approval scripts, and internal automations as model endpoints that business systems call by AgentID.

Share workstation capability

Design, engineering, and data teams can connect office GPU workstations to the public gateway and call specialized local Agents remotely.

Edge and field operations

Factories, labs, and stores can connect outbound to the platform so public systems dispatch tasks without directly entering the field network.

Easier to pass customer review

Customer data, models, and tools stay local by default. With logs off, the platform stores no invocation content or metadata.

Subscribe to the public Agent gateway

Pay for public calling capability: API gateway, tunnel keepalive, AgentID routing, device quota, and access control.

…

Common questions

Quick answers about publishing local Agents as public model endpoints, privacy, and subscriptions.