Built on Claude · Anthropic SDK · No frameworks

Ship AI agents
in days, not months

Three production-ready agent templates with shared infrastructure. Tool calling, memory, cost tracking, and evaluation — all wired up and ready to deploy.

~/my-project
$ pip install starttoagent-kit
✓ Installed 3 agent templates
✓ Core infrastructure ready

$ python -m agents.customer_support
🤖 Agent: "How can I help you today?"
├─ knowledge_base: loaded (47 articles)
├─ escalation: armed
└─ cost_tracker: $0.0000
$
3 TEMPLATES INCLUDED

Agent templates that actually ship

Each template is a complete, tested agent with its own tools, config, and evaluation suite. Fork one and customize, or compose pieces from several.

Most popular
💬

Customer Support

Knowledge base search, sentiment tracking, and automatic escalation to humans. Plug in your docs and go.

  • Knowledge base search
  • Sentiment analysis
  • Human escalation
  • Canned responses
Complex workflows
🔍

Research Assistant

Multi-step research agent that searches, synthesizes, and produces structured reports with source citations.

  • Web search integration
  • Source extraction
  • Report generation
  • Citation tracking
High demand
📄

Document Extraction

Extract structured data from PDFs, invoices, and contracts. Schema validation and confidence scoring built in.

  • PDF parsing
  • Schema validation
  • Confidence scores
  • Batch processing
SHARED INFRASTRUCTURE

The boring parts, done right

Every agent shares a core/ layer of production infrastructure. Thin wrappers over the Anthropic SDK — no framework bloat, no magic, just clean Python.

tools.py 337 lines

Tool Calling

Registry pattern with schema validation, retries, timeouts, and formatted results. Register tools with decorators or Pydantic models.

conversation.py 203 lines

Conversation Management

Message history with sliding window and automatic summarization. Keeps costs bounded without losing context.

cost_tracker.py 230 lines

Cost Tracking

Per-call and per-conversation budgets, automatic pricing lookup, CSV/JSON export. The #1 thing teams wish they'd built earlier.

memory.py 241 lines

Three-Tier Memory

Short-term, session, and long-term (SQLite) storage. Simple key-value API with TTL support and namespace scoping.

errors.py 169 lines

Error Handling

Structured exceptions, Anthropic SDK error mapping, retry decorators with exponential backoff. Claude understands every failure.

eval/ harness + YAML

Evaluation Framework

YAML test cases, automated harness, expected outputs, and budget assertions. Know when your agent breaks before your users do.

PRICING

One purchase. No subscriptions.

Pay once, own it forever. Both plans include the full kit with all three agent templates and shared infrastructure.

Solo Developer

€79 one-time

Everything you need to build and ship AI agents on your own.

  • 3 agent templates (support, research, extraction)
  • Shared core infrastructure
  • Evaluation framework + test cases
  • FastAPI endpoints + Docker config
  • 6 months of updates via GitHub
  • Email support
Get Solo License
Best value

Team License

€149 one-time · up to 5 devs

For teams shipping agents together. Same kit, shared access.

  • Everything in Solo, plus:
  • Up to 5 developer seats
  • 12 months of updates via GitHub
  • Priority email support
  • Team onboarding call (30 min)
  • Private Discord channel
Get Team License

30-day money-back guarantee · No questions asked · Secure checkout via Lemon Squeezy

NEWSLETTER

Get agent-building tips every week

Practical patterns, production gotchas, and new guide announcements. No spam, no AI hype — just things that help you ship.

Join 0 developers · Unsubscribe anytime