Less to build before go-live
The check comes with the framework instead of being written call by call. That's one less project standing between a pilot and production.
SDK 1.0.0
Nobody wants to rebuild a working agent to add controls. With the 1.0.0 release, the check can sit inside the framework your developers already picked, in TypeScript, Python, .NET, Java, Go, or Rust.
Picture an agent that handles refunds. It runs on LangGraph because that's what the team knew, and it works. Then risk asks for a limit: anything over $2,000 needs a manager.
Without support for that framework, a developer has to find every place the agent can issue a refund, wrap each one by hand, and remember to do the same for the next tool someone adds. The rule ends up only as good as the last person who read the code.
An Enforce adapter moves that work into the framework itself. The check runs where the agent already makes its calls, so a new tool or a new step doesn't quietly slip past it.
VeriProof SDK 1.0.0 has a core SDK for each of six languages and 35 Enforce adapters. Every one of them enforces your rules: the action goes ahead, goes ahead with a warning, goes ahead in a corrected form, waits for a person with authority, or stops.
Scroll horizontally to see all columns.
| Language | Core SDK | Enforce adapters for |
|---|---|---|
| TypeScript | @veriproof/sdk-core | Anthropic SDK, Express, Fastify, Genkit, Google GenAI, LangChain.js, LlamaIndexTS, Mastra, Next.js, OpenAI Agents, Vercel AI SDK |
| Python | veriproof-sdk | Anthropic SDK, CrewAI, Google ADK, LangGraph, LlamaIndex, Microsoft Agent Framework, OpenAI Agents, Pydantic AI |
| .NET | Veriproof.Sdk.Core | AutoGen, Microsoft Agent Framework, OpenAI .NET, Prompty, Semantic Kernel |
| Java | com.veriproof:veriproof-sdk-core | LangChain4j, Micronaut LangChain4j, OpenAI Java, Quarkus LangChain4j, Semantic Kernel Java, Spring AI |
| Go | veriproof.app/sdk/go | Anthropic Go, LangChainGo, OpenAI Go |
| Rust | veriproof-sdk | async-openai, Rig |
Each adapter covers a documented part of its framework, such as a model call, a tool call, or a whole run. Anything outside that part stays with the core SDK, and the coverage guide spells out exactly where the line sits for each one.
The check comes with the framework instead of being written call by call. That's one less project standing between a pilot and production.
Your risk team adjusts a limit in VeriProof. Once the change is reviewed and approved, the agent applies it on its next action, without waiting for the next deployment.
When a request falls outside a rule, the agent holds the work and a person with authority decides, with the facts in front of them.
A decision made through CrewAI and one made through Spring AI land in the same history, so an auditor's question has one answer.
Frameworks cover the agents your developers write. They're only part of the picture. Agents built in Copilot Studio and agents that work through a browser answer to the same rules, the same review, and the same decision history.
That's the point we made in the agents you did not write: coverage is a property of your whole estate, not of any one library. SDK 1.0.0 makes the developer-built part of that estate much easier to reach.
A released version never changes. If a fix is needed, it ships as a new patch version with its own release evidence, so a package you tested is the package you run.
Each package was released only after its exact artifact passed a clean install, a security scan, and registry verification, and its bytes are signed. Developers get install commands for their environment from Tools and SDKs in the customer portal, and the SDK 1.0.0 overview covers versions, compatibility, and release notes for each language.
See which frameworks are covered in your team's language, and where the core SDK takes over.