Use Cases
Research agent
Build a research agent that searches, cites, summarizes, and runs as a repeatable job or interactive assistant.
Research agents are where the runtime, tools, and memory layers start to shine together. The common shape is:
- search and browsing
- multi-step synthesis
- source tracking
- optional long-running or scheduled execution
Typical stack
npm install @agentskit/adapters @agentskit/runtime @agentskit/tools @agentskit/memory @agentskit/observabilityRecommended package mix
| Layer | Package | Why it matters |
|---|---|---|
| Provider | @agentskit/adapters | Flex between hosted and local models |
| Runtime | @agentskit/runtime | Lets the agent search, inspect, compare, and iterate |
| Tools | @agentskit/tools | Web search, fetch, browser, filesystem, integrations |
| Memory | @agentskit/memory | Keep prior findings and working context |
| Ops | @agentskit/observability | Compare runs, inspect tool behavior, control cost |
What the architecture usually looks like
- A task arrives as a prompt, CLI command, or scheduled job.
- The runtime decomposes the task into multiple tool-assisted steps.
- Search and browser tools collect evidence.
- The agent writes structured findings to files or downstream systems.
- Observability captures traces for replay and quality review.
Good defaults
- Keep
maxStepsexplicit so research runs stay bounded. - Save source links and snippets as structured output.
- Use memory to avoid re-researching the same topics in follow-up runs.
- Add evals if the research output feeds external customers or exec workflows.
Best follow-up guides
- Build your first agent
- Runtime agent example
- RAG pipeline example
- Research team recipe
- Background agents
- Eval suite
When AgentsKit is especially strong here
Research agents benefit from AgentsKit when you want a clean separation between provider, runtime, tools, and operations. It becomes easy to run the same core workflow in a CLI, a job runner, or a UI without rebuilding the agent from scratch.