For Developers

Build with curate-me.ai

5 SDKs, one platform. Route LLM calls through governance with a base URL swap, build custom agents, orchestrate pipelines, embed chat widgets, and instrument everything.

Full Documentation Try the Dashboard

5-Minute Quickstart

Route all your LLM calls through the governance gateway with zero code changes:

# Install

pip install curate-me

# Swap base URL — that's it

client = OpenAI(

base_url="https://api.curate-me.ai/v1/openrouter/chat/completions",

api_key="cm_org_...",

)

# All calls now have cost tracking, PII scanning, rate limits, and audit trails

SDKs & Libraries

Python SDK

curate-me

pip install curate-me

Docs

Full gateway integration and agent development. Build agents with the BaseAgent pattern, create orchestrated pipelines, and mount as FastAPI endpoints.

Features

Gateway proxy with base URL swap
BaseAgent pattern for custom agents
Pipeline orchestration with dependencies
FastAPI endpoint mounting with SSE
Memory system integration
Cost tracking middleware

Example

from curate_me import CurateMe, BaseAgent

client = CurateMe(api_key="cm_org_...")

# Route any LLM call through the governance gateway
response = client.chat.completions.create(
    model="anthropic/claude-sonnet-4",
    messages=[{"role": "user", "content": "Hello"}],
    # Cost tracking, PII scanning, and rate limits
    # applied automatically by the gateway
)

# Build a custom agent
class ResearchAgent(BaseAgent):
    model = "anthropic/claude-sonnet-4"
    tools = ["web_search", "knowledge_write"]
    budget_per_task = 0.50

    async def execute(self, task: str) -> str:
        return await self.chat(task)

# Create and run a pipeline
pipeline = client.pipeline.create(
    agents=[ResearchAgent(), WriterAgent()],
    flow="sequential",
)
result = await pipeline.run("Research AI governance trends")

TypeScript SDK

@curate-me/sdk

npm install @curate-me/sdk

Docs

Gateway integration for Node.js and browser environments. OpenAI-compatible interface with automatic governance.

Features

OpenAI-compatible chat interface
Streaming with SSE support
Cost headers on every response
Type-safe with full TypeScript types
Browser and Node.js compatible
Middleware pipeline support

Example

import { CurateMe } from "@curate-me/sdk";

const client = new CurateMe({
  apiKey: "cm_org_...",
  orgId: "my-org",
});

// Drop-in replacement for OpenAI — governance applied automatically
const stream = await client.chat.completions.create({
  model: "anthropic/claude-sonnet-4",
  messages: [{ role: "user", content: "Analyze this data..." }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

// Access cost data from response headers
console.log("Cost:", stream.costUsd);
console.log("Tokens:", stream.totalTokens);

CLI

@curate-me/cli

npm install -g @curate-me/cli

Docs

Manage agents, runners, and configuration from the command line. Deploy fleets, check costs, and monitor activity.

Features

Fleet deployment and management
Cost and usage reporting
Agent configuration and testing
Runner lifecycle management
Knowledge base operations
CI/CD pipeline integration

Example

# Login and configure
cm auth login
cm config set org my-org

# Deploy a fleet from template
cm fleet deploy --template development-team \
  --budget 40/day \
  --models "opus,sonnet,gpt-4o"

# Check live costs
cm costs today
# ┌──────────────────┬─────────┬──────────┐
# │ Agent            │ Calls   │ Cost     │
# ├──────────────────┼─────────┼──────────┤
# │ code-review      │ 23      │ $1.47    │
# │ blog-dev         │ 15      │ $0.89    │
# │ docs-agent       │ 8       │ $0.12    │
# └──────────────────┴─────────┴──────────┘

# Monitor agent activity in real-time
cm activity watch --agent blog-dev

# Run guardrails test
cm guardrails test --input "My SSN is 123-45-6789"
# ⚠ PII detected: Social Security Number

Embed Widget

@curate-me/embed

npm install @curate-me/embed

Docs

Drop-in chat widget for any web application. Connects to your agent fleet with full governance — add AI chat in 3 lines of code.

Features

One-line integration for any website
Connects to your fleet agents
Customizable theme and branding
Memory and personalization built in
Cost tracking per user session
White-label support

Example

<!-- Add to any HTML page -->
<script src="https://cdn.curate-me.ai/embed.js"></script>
<script>
  CurateMe.init({
    orgId: "my-org",
    apiKey: "cm_pub_...",    // Public key (safe for browsers)
    agent: "support-agent",   // Which fleet agent to connect
    theme: {
      accent: "#B45309",
      position: "bottom-right",
    },
    memory: true,             // Enable cross-session memory
  });
</script>

<!-- Or with React -->
import { CurateMeChat } from "@curate-me/embed/react";

export function App() {
  return (
    <CurateMeChat
      orgId="my-org"
      agent="support-agent"
      theme={{ accent: "#B45309" }}
    />
  );
}

Observer SDK

@curate-me/observer-sdk

npm install @curate-me/observer-sdk

Docs

Observability and tracing for AI applications. Automatic instrumentation of LLM calls with cost, latency, and quality metrics.

Features

Auto-instrument LLM calls
Distributed trace correlation
Cost and latency metrics
Custom span annotations
OpenTelemetry export
Quality scoring integration

Example

import { Observer } from "@curate-me/observer-sdk";

const observer = new Observer({
  apiKey: "cm_org_...",
  serviceName: "my-app",
});

// Auto-instrument all OpenAI/Anthropic calls
observer.instrument();

// Or manually trace specific operations
const span = observer.startSpan("research-pipeline");
try {
  const result = await researchAgent.run(query);
  span.setAttributes({
    "agent.name": "researcher",
    "agent.cost": result.cost,
    "agent.tokens": result.tokens,
  });
} finally {
  span.end();
}

// View traces in dashboard or export to Jaeger/Datadog
// dashboard.curate-me.ai/traces

Integration Patterns

Gateway Proxy (Simplest)

5 minutes

Swap your base URL. All governance applied automatically. Zero code changes.

Install SDK

Set base URL to gateway

All LLM calls now governed

Agent Builder

30 minutes

Build custom agents with the BaseAgent pattern. Automatic tool calling, memory, and cost tracking.

Define agent class

Deploy to fleet

Pipeline Orchestration

1-2 hours

Chain agents into workflows with dependencies, parallel execution, and conditional routing.

Define pipeline stages

Configure agent routing

Add HITL gates

Deploy

Full Fleet Deployment

Under 1 hour

Deploy a multi-agent team from templates with governance, budgets, and monitoring.

Choose template

Configure per-agent settings

Set governance rules

Deploy fleet

Full API Reference

REST API docs for agents, runners, costs, workflows, knowledge bases, and webhooks.

API Reference

How This Blog Uses the SDK

This blog is a reference application for curate-me.ai. Here's how we integrate:

Gateway Proxy

All LLM calls route through the governance gateway for cost tracking and PII scanning

src/lib/chat/gateway-client.ts

Fleet Orchestrator

Sonnet 4.6 orchestrator delegates to 5 specialist agents with tool calling

src/lib/chat/fleet-orchestrator.ts

Knowledge Base

Dual-write to local PostgreSQL and cloud KB for cross-agent search

src/lib/curate-me.ts

Webhook Handler

Receives agent results, processes refinement loops, sends Slack notifications

src/app/api/webhooks/agent/route.ts

Start Building

Add governance to your AI application in 5 minutes with a base URL swap. Or build a full multi-agent fleet with the SDK.

Documentation Live Demos