Concepts
Concepts
Six contracts. Every AgentsKit package is an implementation of one of them.
Once you have built a real agent, these six contracts explain why the ecosystem composes so cleanly. Every package, every provider, every memory backend, every tool, and every UI binding plugs into one of them.
If you have not done that yet, start with Build your first agent and come back here after you have seen the stack in motion.
#The six contracts
| Contract | What it abstracts | Page |
|---|---|---|
| Adapter | LLM providers | adapter |
| Tool | callable functions | tool |
| Skill | personas + prompts | skill |
| Memory | conversation + vector state | memory |
| Retriever | context retrieval | retriever |
| Runtime | the orchestration loop | runtime |
#Start here
- Mental model — how the six contracts compose.
- Implementations — every concrete implementation per contract.
- Errors — the error taxonomy every contract honors.
#See also
- Packages overview — packages grouped by role.
- For agents — dense LLM-friendly reference per contract.
- Comparison — AgentsKit vs LangChain / Vercel AI / Mastra.
Explore nearby
- PeerMental model
The six concepts every AgentsKit user should know.
- PeerAdapter
The seam between AgentsKit and an LLM provider. The same interface for OpenAI, Anthropic, Gemini, Ollama, or anything that streams tokens.
- PeerTool
A function the model can call, with JSON-schema-typed arguments and a JSON-serializable result.