Supported agents

Connect agents however they were built

Framework, custom code, Microsoft Copilot Studio, MCP, or a browser: however an agent was built, VeriProof checks its actions against your rules and records each decision the same way.

How was your agent built?

Find the closest match. Many teams use more than one.

  1. On an agent framework such as LangGraph, CrewAI, or the OpenAI Agents SDK

    Connects with
    A ready-made VeriProof adapter for that framework
    VeriProof checks
    Requests to the AI model, use of tools, or a whole task, depending on the framework
    Find your framework
  2. In your own code, without a framework

    Connects with
    The VeriProof developer kit for TypeScript, Python, .NET, Java, Rust, or Go
    VeriProof checks
    Any AI request, tool use, or business action your developers choose
    Open the SDK quickstart
  3. In Microsoft Copilot Studio

    Connects with
    VeriProof added to the agent, plus a check in the system that makes the change
    VeriProof checks
    Business actions, such as updating a loan, before the record changes
    See Copilot Studio
  4. As an AI assistant connected through MCP (Model Context Protocol)

    Connects with
    VeriProof’s MCP connection
    VeriProof checks
    Proposed actions, limited to what the signed-in person may do
    See MCP
  5. As an agent that clicks through a browser or desktop app

    Connects with
    VeriProof’s browser-agent tools, through MCP
    VeriProof checks
    The proposed click or entry, recorded with the screen it was based on
    See computer-use governance
  6. As an agent that pays with x402, pay.sh, AP2, or stablecoins

    Connects with
    The VeriProof developer kit, at the point where the agent pays
    VeriProof checks
    Each payment, before the wallet or provider moves money
    See agent payments
  7. On a platform with no VeriProof SDK

    Connects with
    The VeriProof API, plus a command-line tool for setup and exports
    VeriProof checks
    Any action the platform sends to VeriProof before carrying it out
    See the API reference

Framework adapters by language

An adapter plugs into the agent framework your developers already use, so they don't have to add the check by hand at every step. 35 adapters are available today.

Scroll horizontally to see all columns.

LanguageCore SDKEnforce adapters for
TypeScript@veriproof/sdk-coreAnthropic SDK, Express, Fastify, Genkit, Google GenAI, LangChain.js, LlamaIndexTS, Mastra, Next.js, OpenAI Agents, Vercel AI SDK
Pythonveriproof-sdkAnthropic SDK, CrewAI, Google ADK, LangGraph, LlamaIndex, Microsoft Agent Framework, OpenAI Agents, Pydantic AI
.NETVeriproof.Sdk.CoreAutoGen, Microsoft Agent Framework, OpenAI .NET, Prompty, Semantic Kernel
Javacom.veriproof:veriproof-sdk-coreLangChain4j, Micronaut LangChain4j, OpenAI Java, Quarkus LangChain4j, Semantic Kernel Java, Spring AI
Goveriproof.app/sdk/goAnthropic Go, LangChainGo, OpenAI Go
Rustveriproof-sdkasync-openai, Rig

See what each adapter covers →

Agents that spend money

An agent that spends money needs a limit, and a person who approves the exceptions. VeriProof checks each payment before your wallet or provider moves any money, and sends anything over the limit to the budget owner.

Pay-per-use data and tools (x402)

The agent pays a small amount each time it uses a paid data source or tool. Your rules cap what it spends and which payment networks it can use.

See x402 →

pay.sh and Machine Payments Protocol

The agent picks a paid provider from a catalog. Rules check that the provider is verified and that the session stays under its cap.

See pay.sh →

Purchases on someone's behalf (AP2)

The agent checks out for a customer or employee. Your rules hold it to the payee, amount, and time limit that person approved.

See AP2 →

Stablecoin supplier payments

The agent pays suppliers in stablecoins. Payments to new suppliers and large cross-border transfers go to your treasury team first.

See stablecoin payments →

Same record, six portals

Decisions from every agent reach the right portal

However an agent connects, its decisions land in portals built for the business owner, risk and compliance, governance engineers, developers, auditors, and administrators. Work moves between them through built-in handoffs.

  • An agent action needs a person

    Reviewer with authority

  • A rule needs data the agent does not send

    Developer

  • A governance engineer changes a rule for a production agent

    Administrator or Compliance Officer

  • A control is not met

    Assigned owner

  • A risk needs a business decision

    Business Owner

  • An audit begins

    Auditor

Looking for Slack, ServiceNow, or Vanta?

Those connect the tools your teams work in rather than your agents, so they have their own directory, organized by human-in-the-loop reviews, GRC and compliance, and security and monitoring. Browse integrations →

Running an agent on something we don't list?

Tell us where it runs and what it needs to do. We'll tell you which connection fits and reply within one business day. If you build agents for clients, the partner program covers implementation work.