Provider-agnostic
One LlmProvider trait, five first-party implementations. Switch models by
swapping a config line — your tools and loop stay identical.
Kova SDK
An async-first Rust library for building LLM-powered agents. Provider-agnostic, trait-based, production-ready.
Build agents, not plumbing
Section titled “Build agents, not plumbing”Kova gives you one clean, trait-based core — Agent, LlmProvider, Tool —
and lets you swap the pieces underneath. Point it at Anthropic, OpenAI, Bedrock,
Gemini, or a local Ollama model without touching your agent code.
Tool calling built in
Implement the Tool trait, register it, done. The agent runs the
call-execute-observe loop for you, with concurrent tool execution.
Stateless by design
Agent::run takes your history in and hands the new messages back — you own
persistence, sessions, and scaling. No hidden state.
Structured output
Constrain a turn’s final answer to a JSON schema and parse it straight into
your type with run_structured::<T>() — mapped natively per provider.
Streaming & thinking models
Pull-based streams of text, tool calls, and chain-of-thought — the same API across every provider.
Prompt caching & embeddings
Automatic prompt caching on Anthropic and Bedrock, with per-turn cache
accounting. Plus an EmbeddingProvider seam for host-side retrieval.
MCP integration
Connect Model Context Protocol servers over stdio, HTTP+SSE, or Streamable HTTP with OAuth — with auto-reconnect. Remote tools look like native ones.
Reliable & cancellable
Automatic retries with backoff, classified provider errors, context-budget
guards, and CancellationToken-driven turn cancellation.
Where to next
Section titled “Where to next” Introduction What Kova is, who it's for, and the mental model behind it.
Installation Add the crate, pick your feature flags.
Quick Start A running agent in under five minutes.
The Agentic Loop The single most important concept in the SDK.