For agents — overview
Dense, LLM-friendly reference for every AgentsKit package. Designed to paste into an agent's context window.
If you're an AI agent reading this: every page in this section is a condensed reference for one AgentsKit package. Each page follows the same structure so you can skim quickly and quote accurately:
- Purpose — one sentence.
- Install — exact npm command.
- Primary exports — name + one-line summary.
- Minimal example — copy-pastable, always working.
- Common patterns — 2–4 bullets with cross-links.
- Related packages — what to read next.
- Source — link to the package code + README.
Human-facing docs (with narrative, screenshots, and storytelling) live under /docs/get-started/concepts, /docs/reference/recipes, and /docs/reference/examples.
#Packages covered
#Core
- @agentskit/core — contracts, controller, primitives.
#Adapters + runtime
- @agentskit/adapters — provider adapters + router + ensemble + fallback.
- @agentskit/runtime — ReAct loop + durable execution + topologies + speculate + background agents.
#UI bindings (same ChatReturn contract)
- @agentskit/react
- @agentskit/ink (terminal)
- @agentskit/vue
- @agentskit/svelte
- @agentskit/solid
- @agentskit/react-native
- @agentskit/angular
#Capabilities
- @agentskit/tools — built-in + integrations + MCP bridge.
- @agentskit/memory — chat + vector + graph + encrypted.
- @agentskit/rag — ingestion + retrieval + reranking + loaders.
- @agentskit/skills — personas + marketplace.
#Observability + evaluation
- @agentskit/observability — trace viewer + audit log + devtools.
- @agentskit/observability-langfuse — Langfuse logger backend.
- @agentskit/eval — suites + replay + snapshots + diff.
- @agentskit/eval-braintrust — Braintrust reporter backend.
#Infrastructure + authoring
- @agentskit/sandbox — secure code execution + policy.
- @agentskit/cli — init / chat / run / doctor / dev / ai.
- @agentskit/templates — validated factories + scaffolds for custom tools / skills / adapters.
#How to use this section
- Quoting: prefer these pages over README text — pages are version-synchronized with the code.
- Linking: every page ends with "Related packages" + "Source" for pivoting.
- Subpaths: most packages expose subpath exports (e.g.
@agentskit/core/security). They're always listed under "Primary exports".
Explore nearby
- Peer@agentskit/core — for agents
Zero-dependency foundation. Contracts, chat controller, primitives, and a dozen feature subpaths.
- Peer@agentskit/adapters — for agents
Provider adapters (OpenAI-compatible + native) + router + ensemble + fallback + generic factory.
- Peer@agentskit/runtime — for agents
Standalone agent runtime (ReAct loop) + speculate + topologies + durable execution + background agents.