Switch providers without rewriting your request.

The LLM client Aktagon built to power its agents, in Go, TypeScript, Python, and Rust. One API for Anthropic, OpenAI, Google, and 20+ more, including local models through Ollama and vLLM.

anthropicopenaigooglebedrockollamavllm 20+ more

Go TypeScript Python Rust

c := llmkit.New("anthropic", key)

resp, err := c.Text.
    System("You plan sprints.").
    Prompt(ctx, "Plan the next sprint.")

const client = anthropic(process.env.ANTHROPIC_API_KEY);

const resp = await client.text
  .system("You plan sprints.")
  .prompt("Plan the next sprint.");

c = anthropic(os.environ["ANTHROPIC_API_KEY"])

resp = await (
    c.text
    .system("You plan sprints.")
    .prompt("Plan the next sprint.")
)

let c = anthropic(key);

let resp = c
    .text()
    .system("You plan sprints.")
    .prompt("Plan the next sprint.")
    .await?;

Text generationChatCompletion

The core call. Same builder chain in every language.

StreamingChatCompletion · stream

Token-by-token output; the final response carries usage.

Reasoning effortReasoning

Ask the model to think harder; usage reports reasoning tokens.

Prompt cachingCaching

Reuse a large system prefix; usage reports cache read/write. OpenAI automatic, Anthropic explicit, Google resource.

Batch promptingBatching

Many prompts, one call. Anthropic, OpenAI.

Agent tool loopToolCalling

Model catalogueCatalogue

Introspect models and providers, filter by capability, list live.

Image generationImageGeneration

Generate and image-to-image (vision). Providers: google, grok, openai, openrouter, together, vertex.

Middleware (observe + veto)Middleware

Hooks fire pre/post each op; returning an error vetoes the call.

File uploadFileUpload

Two forms: a path, or raw bytes with an explicit filename + mime.

Music generationMusicGeneration

Vertex Lyria 2, instrumental, ~30s WAV. Providers: google, minimax, vertex.

Video generationVideoGeneration

Asynchronous: submit returns a handle, wait() polls to a hosted URL. Provider: grok.