ServiceWeave

ServiceWeave is an AI-driven Kubernetes operator that implements an Agentic Mesh architecture. It enables microservices to be augmented with LLM-powered Service Agent sidecars, allowing services to understand natural language requests and automatically invoke appropriate APIs.

What is ServiceWeave?

ServiceWeave works similarly to how Istio injects Envoy proxies into your pods, but instead injects intelligent agents capable of semantic understanding. These agents can:

• Understand natural language requests and translate them into API calls
• Discover API schemas automatically (OpenAPI, GraphQL, gRPC)
• Cache intent-to-tool mappings for improved performance
• Implement risk-based approval workflows for sensitive operations

Key Features

Feature	Description
Transparent Injection	Automatic sidecar injection via Kubernetes mutating webhooks
Multi-Schema Support	Works with OpenAPI 3.x, GraphQL, and gRPC services
Flexible LLM Integration	Compatible with any OpenAI-compatible API provider
Risk Management	Four-tier risk classification (Autonomous, Notify, Approve, Delegate)
Semantic Caching	Intent and response caching with configurable TTLs
Observability	Built-in OpenTelemetry integration for distributed tracing

How It Works

flowchart TB
    subgraph cluster["Kubernetes Cluster"]
        subgraph operator["ServiceWeave Operator"]
            mc["MeshConfig<br/>Controller"]
            sac["ServiceAgent<br/>Config Controller"]
            pm["Pod Mutator<br/>(Webhook)"]
        end
        subgraph pod["Your Application Pod"]
            app["Your App<br/>Container"]
            agent["ServiceWeave Agent<br/>(Injected Sidecar)"]
            app <--> agent
        end
    end

Loading

1. MeshConfig defines global LLM, embedding, and vector store settings
2. ServiceAgentConfig configures per-service API schema discovery
3. Pod Mutator automatically injects the agent sidecar into labeled namespaces
4. ServiceWeave Agent handles natural language requests and API orchestration

Quick Example

Enable ServiceWeave in a namespace:

kubectl label namespace my-app serviceweave.ai/inject=enabled

Create a global mesh configuration:

apiVersion: serviceweave.ai/v1
kind: MeshConfig
metadata:
  name: serviceweave
spec:
  llm:
    baseURL: "https://api.openai.com/v1"
    defaultModel: "gpt-4o-mini"
    apiKeySecretRef:
      name: openai-credentials
      key: api-key
  embedding:
    baseURL: "https://api.openai.com/v1"
    model: "text-embedding-3-small"
    apiKeySecretRef:
      name: openai-credentials
      key: api-key
  vectorStore:
    type: qdrant
    endpoint: "qdrant.serviceweave-system.svc:6333"
  injection:
    namespaceSelector:
      matchLabels:
        serviceweave.ai/inject: "enabled"

Configure a service for agent discovery:

apiVersion: serviceweave.ai/v1
kind: ServiceAgentConfig
metadata:
  name: order-service
  namespace: my-app
spec:
  schema:
    schemaPath: "/openapi.json"
    schemaType: openapi
    description: "Order management service for creating, updating, and tracking orders"
  riskTier: 0  # Autonomous execution

Next Steps

• Quick Start - Get up and running in minutes
• Architecture - Understand the system design
• MeshConfig Reference - Full configuration options