Files · OpenAI Cost Control for SMB Agent Workflows

# OpenAI Cost Control for SMB Agent Workflows
 
> Slash OpenAI spending with AI-powered caching and budget enforcement for small business support teams.
 
A tutorialized reference solution from [reaatech.com](https://reaatech.com), demonstrating how to build production-grade AI cost control with the `@reaatech/*` package family.
 
## Architecture
 
This solution wraps every OpenAI call with `@reaatech/llm-cost-telemetry` for real-time token cost capture, enforces monthly budgets via `@reaatech/agent-budget-engine`, and uses `@reaatech/llm-cache` with Redis to serve cached responses for semantically similar prompts, cutting redundant API costs. `@reaatech/llm-cost-telemetry-aggregation` rolls up spend across tenants, and Langfuse provides OTel-compliant observability.
 
## API Endpoints
 
| Endpoint | Method | Description |
|---|---|---|
| `/api/cost?tenantId=X` | GET | Aggregated spend data per tenant |
| `/api/budget?tenantId=X` | GET | Current budget state and disabled tools |
| `/api/budget` | POST | Pre-flight budget check (`{ tenantId, estimatedCost, modelId }`) |
| `/api/usage?tenantId=X` | GET | Usage history for a tenant |
| `/api/cache` | GET | Cache health status |
 
## Environment Variables
 
| Variable | Description |
|---|---|
| `OPENAI_API_KEY` | OpenAI API key for LLM calls |
| `REDIS_URL` | Redis connection URL for cache backend |
| `DATABASE_URL` | PostgreSQL connection string for settings store |
| `LANGFUSE_PUBLIC_KEY` | Langfuse public key for observability |
| `LANGFUSE_SECRET_KEY` | Langfuse secret key for observability |
| `LANGFUSE_HOST` | Langfuse host URL |
| `DEFAULT_DAILY_BUDGET` | Default daily budget per tenant (USD) |
| `DEFAULT_MONTHLY_BUDGET` | Default monthly budget per tenant (USD) |
 
## Quick Start
 
1. Create a PostgreSQL database
2. Copy `.env.example` to `.env` and fill in your credentials
3. Start Redis on localhost:6379
4. Run `pnpm install`
5. Run `pnpm dev`
 
## Running Tests
 
```bash
pnpm test
```
 
## Project Layout
 
```
app/
  api/
    cost/route.ts     Aggregated spend endpoint
    budget/route.ts    Budget check and state endpoints
    usage/route.ts     Usage history endpoint
    cache/route.ts     Cache health endpoint
src/
  lib/
    cost-types.ts         Shared type definitions
    db.ts                 Database connection
    db/schema.ts          Drizzle ORM schema
    db/index.ts           Database barrel exports
    wrap-openai.ts        OpenAI SDK wrapper with telemetry
    budget-middleware.ts  Budget enforcement via agent-budget-engine
    cache-layer.ts        Semantic caching with Redis
    aggregation.ts        Cost collection and aggregation
    langfuse.ts           Langfuse observability
    settings-store.ts     Drizzle-backed tenant settings
    orchestrator.ts       Unified pipeline orchestrator
tests/
  lib/                    Test suite mirroring src/lib/
  app/api/                Integration tests for API routes
```
 
## License
 
MIT — see [LICENSE](./LICENSE).