Skip to main content
Every adapter wraps your framework’s normal call surface and inserts a check() step that returns an HTTP 402 challenge when payment is missing. The client pays on-chain, retries with proof, and the call goes through. Same shape across all 12 frameworks.

​
Available adapters

FrameworkWhat it gatesSource
LangChainLLMs, chains, agent tool endpointslangchain
LangGraphCompiled StateGraph executionlanggraph
CrewAICrews, tasks, agent toolscrewai
AutoGenConversations, callable toolsautogen
AG2Multi-agent flows(via AutoGen)
DSPyModules, programs, chainsdspy
LlamaIndexLLMs, query engines, chat engines, ReAct agent toolsllamaindex
Pydantic AIAgents, model callspydantic-ai
Vercel AI SDKModel calls, tool callsvercel-ai-sdk
Semantic KernelChat completions, KernelFunction, pluginssemantic-kernel
Hugging FaceInferenceClient, transformers pipelines, smolagents toolshuggingface
AgnoAgno agentsagno
Google A2AA2A endpoints (server side)a2a

​
Common shape

Every adapter has the same three-method surface: 
from algovoi_<framework> import AlgoVoi<Framework>

adapter = AlgoVoi<Framework>(
    api_base="https://api1.ilovechicken.co.uk",
    api_key="algv_…",
    tenant_id="…",
    webhook_secret="algvw_…",
)

# 1. Check payment status before invocation
status = adapter.check(resource_id="premium-call")

# 2. If unpaid, surface a challenge to the client
if not status.paid:
    raise adapter.PaymentRequired(status.challenge)

# 3. Run the wrapped call
result = adapter.run(...)  # framework-specific

​
Picking a protocol

Each adapter supports x402, MPP, and AP2 on the same code path. Pick the one that matches your client:
  • x402 for HTTP-fronted endpoints. The most common case for LangChain or Vercel AI SDK servers.
  • MPP for JSON-RPC or MCP-server-backed tools. Use when the call is method-shaped, not URL-shaped.
  • AP2 for autonomous agents acting on a signed mandate.
See the protocol pages for detail: x402, MPP, AP2.

​
Versioning

All adapters track v1.0.0 of the shared client surface. Behaviour and method names are identical across frameworks; only the wrapping layer changes. If you switch frameworks you keep the same AlgoVoi configuration.