| Layer | What it does |
|---|---|
| Agent | The execution loop. Model calls, tool orchestration, structured outputs, context management, middleware, and tracing |
| Capabilities | Tools, knowledge (RAG), memory, skills, guardrails, security, evaluation |
| Interfaces | Deploy agents to Telegram, Discord, Slack, WhatsApp, email, voice calls, WebSocket, or custom platforms |
| Runtime | HTTP endpoints, cron scheduling, webhooks, multi-interface serving, and hot-reload dev mode |
What You Can Build
agent.py
Start Building
- Your First Agent Build and run an agent with tools in under 20 lines of code.
- Agents The core execution loop: models, tools, and instructions.
- Teams Coordinate multiple agents to collaborate, route, or divide work.
- Workflows Orchestrate agents through sequential, parallel, conditional, and iterative steps.
- Models 10 providers, string shorthand, streaming, structured output, vision, and audio.
- Knowledge Full RAG pipeline: readers, chunkers, embedders, rerankers, vector databases.
Built for Production
- Async-first with sync wrappers for every API.
- 10 LLM providers with automatic key rotation and failover.
- Token counting, cost tracking, and usage analytics.
- Input/output/tool guardrails and security hardening.
- Native tracing with JSONL and debug exporters.
- Full type annotations with
py.typedmarker for mypy consumers.
AI agents require composable architecture. Definable is the framework for building it.