AI Agent Framework
AI Agent Framework is a TypeScript framework for building AI chains and tool-using agents from composable primitives.
Key Capabilities
- runnable composition for deterministic workflows
- prompt templates and output parsers
- model adapters for provider integration
- agent runtime with tool execution
- resumable
Workflowprimitive with step snapshots - typed building blocks for application code
When To Use It
Use AI Agent Framework when you need to:
- build structured prompt/model/parser pipelines
- compose reusable AI workflow stages
- add tool calling to an assistant
- keep provider integration separate from application flow
Core Concepts
| Concept | Meaning |
|---|---|
| Runnable | executable composable unit |
| Chain | fixed sequence of runnables |
| Prompt template | formatter for prompt variables |
| Output parser | converter from model text to application output |
| Agent | runtime that can call tools before returning a response |
Current Scope
The current MVP includes runnable composition, prompt templates, parsers, OpenAI provider integration, workflow snapshots/resume state, and a tool-capable agent runtime. Retriever primitives, richer memory support, and additional provider adapters remain planned work.
Documentation
- Getting Started: install and run the first chain and agent.
- User Guide: choose primitives for common use cases.
- Examples: practical usage patterns.
- API Reference: public API surface.
- Architecture: runtime behavior.
- Production Readiness: hardening checklist and rollout guidance.