One programming model for building with LLMs across TypeScript, Python, Java, C++, Go, and Rust.

Ax is TypeScript-first and ships today as @ax-llm/ax. The same signatures, provider mappings, agents, flows, runtime contracts, and optimizers are also compiled into verified generated Python, Java, C++, Go, and Rust libraries.

• Signatures for typed structured generation: string DSL, fluent f() builder, or any Standard Schema v1 validator — Zod, Valibot, ArkType.
• Provider abstraction across OpenAI-compatible endpoints, OpenAI Responses, Anthropic, Gemini, Grok/xAI, Mistral, Cohere, Reka, DeepSeek, Azure OpenAI, audio, and realtime event streams.
• Agents with runtime execution, context budgets, checkpoints, action-log replay, discovery, memory, skills, and delegation.
• Flows as typed program graphs with branches, loops, feedback, cache behavior, parallel execution, and .returns(...) projection.
• Optimizers including GEPA, few-shot bootstrapping, portable optimizer artifacts, and evaluation/apply flows.
• One semantic core compiled into TypeScript, Python, Java, C++, Go, and Rust library shapes, so the same Ax program model can move across runtime stacks.

flowchart LR
  S["Signature (string, f, Standard Schema)"] --> G["AxGen typed generation"]
  G --> P["Provider descriptors / AI clients"]
  G --> A["AxAgent"]
  G --> F["AxFlow"]
  G --> O["GEPA / optimizer artifacts"]
  C["Shared Ax semantics"] --> TS["TypeScript"]
  C --> PY["Python"]
  C --> JV["Java"]
  C --> CP["C++"]
  C --> GO["Go"]
  C --> RS["Rust"]

The TypeScript package is the source implementation and the current published package:

import { ai, ax } from "@ax-llm/ax";

const llm = ai({ name: "openai", apiKey: process.env.OPENAI_APIKEY });

const classify = ax(
  'review:string -> sentiment:class "positive, negative, neutral"',
);

const { sentiment } = await classify.forward(llm, {
  review: "This product is amazing!",
});
// sentiment: "positive" — typed as the literal union

No prompt engineering. Switch name: "openai" to "anthropic", "google-gemini", "mistral", "deepseek", "grok", etc. — same signature, same code.

The generated Python, Java, C++, Go, and Rust libraries expose the same top-level Ax ideas in native package shapes. Their generated source is checked in under packages/<language> so the supported APIs are easy to inspect. The repo runner uses those committed packages and runs examples without asking you to remember compiler commands:

npm run example -- python signature_schema.py
npm run example -- java SignatureSchemaExample.java
npm run example -- cpp signature_schema.cpp
npm run example -- go signature_schema.go
npm run example -- rust signature_schema.rs

See src/examples/README.md for runnable examples, docs/RELEASE.md for package/release shape, and docs/COMPILER.md for how the language-agnostic Ax compiler works. When AxIR changes, run npm run axir:generate-packages to refresh the checked-in packages.

Ax is designed to stay in the same latency class as direct provider calls while adding typed outputs, validation, retries, tools, tracing, and memory. The hot path is intentionally thin: render the signature, call the provider, parse the result, and return a typed value.

Streaming is the default because it lets Ax do useful work before the model finishes: parse fields as they arrive, run streaming assertions, fail early, cancel the in-flight stream, and start correction without spending tokens on an output that is already known to be invalid. When you only want a final object, forward() still gives you one; when you want incremental output, streamingForward() exposes the stream directly.

The repo includes a streaming benchmark for checking overhead on your own providers and models:

AX_STREAM_BENCH_PROVIDER=anthropic AX_STREAM_BENCH_MODEL=claude-sonnet-4-5-20250929 AX_STREAM_BENCH_RUNS=2 AX_STREAM_BENCH_WARMUP_RUNS=0 npm run tsx src/examples/streaming-latency.ts
AX_STREAM_BENCH_PROVIDER=google-gemini AX_STREAM_BENCH_MODEL=gemini-2.5-flash AX_STREAM_BENCH_RUNS=2 AX_STREAM_BENCH_WARMUP_RUNS=0 npm run tsx src/examples/streaming-latency.ts

Recent runs on Claude Haiku/Sonnet and Gemini Flash/Flash Lite show provider queueing and model generation dominate total latency; AxGen stays close to the raw ai.chat() path while providing the structured-output control loop that direct SDK calls leave to application code.

const extract = ax(`
  customerEmail:string, currentDate:datetime ->
  priority:class "high, normal, low",
  sentiment:class "positive, negative, neutral",
  ticketNumber?:number,
  nextSteps:string[],
  estimatedResponseTime:string
`);

const result = await extract.forward(llm, {
  customerEmail: "Order #12345 hasn't arrived. Need this resolved immediately!",
  currentDate: new Date(),
});

import { ax, f } from "@ax-llm/ax";

const productExtractor = f()
  .input("productPage", f.string())
  .output("product", f.object({
    name: f.string(),
    price: f.number(),
    specs: f.object({
      dimensions: f.object({ width: f.number(), height: f.number() }),
      materials: f.array(f.string()),
    }),
    reviews: f.array(f.object({ rating: f.number(), comment: f.string() })),
  }))
  .build();

const gen = ax(productExtractor);
const { product } = await gen.forward(llm, { productPage: "..." });
// product.specs.dimensions.width is typed end-to-end

Any Standard Schema v1 validator works wherever f.* is accepted — at field level, whole-object level, or on a fn() tool. Same retry pipeline, same type inference, no adapter.

import { z } from "zod";
import { ax, f, fn } from "@ax-llm/ax";

// (1) Per-field zod — mix freely with f.* fields
const reviewSentiment = ax(
  f()
    .input("productName", z.string().describe("Reviewed product"))
    .input("reviewText", z.string().min(10))
    .output("sentiment", z.enum(["positive", "neutral", "negative"]))
    .output("score", z.number().min(1).max(10))
    .output("keyPoints", z.array(z.string()))
    .build(),
);

// (2) Whole-object zod — declare once, decomposed into ordered fields
const productSummary = ax(
  f()
    .input(z.object({ productName: z.string(), buyerProfile: z.string() }))
    .output(z.object({
      headline: z.string(),
      pros: z.array(z.string()),
      cons: z.array(z.string()),
      recommendation: z.enum(["buy", "wait", "skip"]),
    }))
    .build(),
);

// (3) Whole-object zod on fn() — typed tool definition
const lookupProduct = fn("lookupProduct")
  .description("Look up a product by name")
  .arg(z.object({ productName: z.string().min(1), includeSpecs: z.boolean().optional() }))
  .returns(z.object({ price: z.number(), inStock: z.boolean(), rating: z.number().min(1).max(5) }))
  .handler(async ({ productName }) => ({ price: 79.99, inStock: true, rating: 4.3 }))
  .build();

.min(), .max(), .email(), .url(), .regex() feed the normal retry pipeline; .refine(), .transform(), and .superRefine() execute at parse time on complete field values, in both streaming and non-streaming. Cache breakpoints and internal reasoning fields use companion options: { cache: true }, { internal: true }. Multimodal inputs (image, audio, file) still use f.*.

Runnable: src/examples/standard-schema.ts.

const assistant = ax("question:string -> answer:string", {
  functions: [
    { name: "getCurrentWeather", func: weatherAPI },
    { name: "searchNews", func: newsAPI },
  ],
});

const { answer } = await assistant.forward(llm, {
  question: "What's the weather in Tokyo and any news about it?",
});

const analyze = ax(`
  image:image, question:string ->
  description:string,
  mainColors:string[],
  category:class "electronics, clothing, food, other",
  estimatedPrice:string
`);

Batch speech APIs are exposed by AI services: ai.transcribe({ audio }) turns audio into text, and ai.speak({ text }) turns text into an audio artifact. Signature audio outputs are scripted artifacts: the model writes the text for speech:audio, then Ax synthesizes it after parsing.

const say = ax("question:string -> speech:audio, summary:string");
const res = await say.forward(llm, { question: "Greet the team." }, {
  speech: { speak: { voice: "alloy", format: "mp3" } },
});

console.log(res.speech.data);       // base64 audio
console.log(res.speech.transcript); // generated script

Agents transcribe :audio inputs before the planner/executor/responder stages, so tools and memory receive stable text rather than base64 payloads. Native conversational audio is still available through .chat().

OpenAI supports both request-based audio chat (gpt-audio, gpt-audio-mini) and realtime voice/transcription models (gpt-realtime-2, gpt-realtime-whisper). Gemini native audio uses the Live API under the same .chat() shape; Grok Voice uses the realtime voice endpoint.

import WebSocket from "ws";
import {
  ai,
  axAIOpenAIRealtimeDefaultConfig,
  axAIOpenAIRealtimeTranscriptionDefaultConfig,
} from "@ax-llm/ax";

const voice = ai({
  name: "openai",
  apiKey: process.env.OPENAI_APIKEY!,
  config: axAIOpenAIRealtimeDefaultConfig(), // gpt-realtime-2
});

const stream = await voice.chat(
  { chatPrompt: [{ role: "user", content: "Say hello out loud." }] },
  { stream: true, webSocket: WebSocket },
);

for await (const chunk of stream) {
  const audio = chunk.results[0]?.audio;
  if (audio?.isDelta) {
    // base64 pcm16 audio bytes
    process.stdout.write(".");
  }
}

const transcriber = ai({
  name: "openai",
  apiKey: process.env.OPENAI_APIKEY!,
  config: axAIOpenAIRealtimeTranscriptionDefaultConfig(), // gpt-realtime-whisper
});

Runnable: src/examples/audio-chat.ts streams realtime audio, saves a WAV, and plays it when a local player is available. src/examples/audio-batch-and-agent.ts writes generated MP3 artifacts under src/examples/output/ and plays them immediately.

AxAgent is a three-stage pipeline that turns a signature into a long-running, tool-using actor. Each forward() call runs distiller → executor → responder.

flowchart LR
  IN["inputs"] --> D["Distiller"]
  D --> E["Executor (RLM loop)"]
  E --> RT["AxJSRuntime sandbox"]
  E --> FN["functions / child agents"]
  E --> M["recall - memories"]
  E --> SK["consult - skills"]
  E --> RES["Responder"]
  RES --> OUT["typed output"]

import { agent, AxJSRuntime } from "@ax-llm/ax";

const analyzer = agent(
  "context:string, query:string -> answer:string, evidence:string[]",
  {
    agentIdentity: {
      name: "documentAnalyzer",
      description: "Analyze long documents with iterative code + sub-queries",
    },
    contextFields: ["context"],
    runtime: new AxJSRuntime(),
    maxTurns: 20,
    maxRuntimeChars: 2_000,
    contextPolicy: { preset: "checkpointed", budget: "balanced" },
    executorOptions: { model: "gpt-4o-mini" },
  },
);

const result = await analyzer.forward(llm, {
  context: veryLongDocument,
  query: "What are the main arguments and supporting evidence?",
});

The recursive runtime (RLM) keeps long context out of the root prompt: the executor runs JS in a persistent sandboxed session, narrows context with llmQuery(...) sub-calls, and uses checkpointed replay so older turns collapse into summaries instead of growing the prompt unbounded.

Runnable: src/examples/rlm-agent-controlled.ts, src/examples/rlm-discovery.ts.

Four orthogonal options on agent(...). Opt in to what the task needs.

Context map — a small persistent orientation cache for repeated questions over the same long context. When configured, Ax shows it to the distiller and updates it once after each successful completed run. By default the map keeps evolving forever; set infiniteEvolve: false with evolveSteps on the map object to do a finite warmup and then reuse a frozen map. Use onUpdate to save the new snapshot wherever your app stores state.

import { agent, AxAgentContextMap } from "@ax-llm/ax";

const map = new AxAgentContextMap(savedSnapshot, {
  maxChars: 4000,
  infiniteEvolve: false,
  evolveSteps: 10,
});

const analyzer = agent("context:string, query:string -> answer:string", {
  contextFields: ["context"],
  contextMap: {
    map,
    onUpdate: ({ map }) => saveSnapshot(map.snapshot()),
  },
});

Memories — vector / BM25 / KV lookup the actor controls via await recall([...]). Results land on inputs.memories for the next turn. Lifetime is one .forward(); persist externally to carry across calls.

const myAgent = agent("task:string -> plan:string", {
  onMemoriesSearch: async (searches, alreadyLoaded) => {
    const skip = new Set(alreadyLoaded.map((m) => m.id));
    return (await myVectorDB.searchBatch(searches, { topK: 3 }))
      .filter((m) => !skip.has(m.id));
  },
  onUsedMemories: (results) => console.log("[memories]", results.map((r) => r.id)),
});

Skills — guidance / runbook bodies the actor pulls in on demand via await consult([...]). Loaded skills render under "Loaded Skills" in the executor system prompt and persist across .forward() calls.

const myAgent = agent("task:string -> plan:string", {
  onSkillsSearch: async (searches) =>
    mySkillStore.searchBatch(searches, { topK: 2 }),
  // Or preload statically — `consult()` not required:
  skills: [{ name: "release-checklist", content: "1. Bump version\n2. ..." }],
});

Sandboxed JS runtime — AxJSRuntime is the default; it is hardened by default and portable across Node, Bun (smol: true workers), Deno, and the browser. Capabilities are opt-in via permissions.

import { AxJSRuntime, AxJSRuntimePermission } from "@ax-llm/ax";

const runtime = new AxJSRuntime({
  permissions: [AxJSRuntimePermission.NETWORK], // grant fetch only
});

Defaults: import() blocked, intrinsics frozen, ShadowRealm locked, worker IPC locked, and on Node 20+ the OS Permission Model auto-engages as a second defense layer. Add FILESYSTEM, STORAGE, CHILD_PROCESS, etc. only as the task requires.

Security model: the runtime is defense-in-depth for LLM-authored code, not a container or VM boundary. Host callbacks and the permissions you grant remain the authority boundary; keep durable secrets and privileged effects in host-side functions.

Runnable: src/examples/rlm-memories-and-skills.ts.

AxFlow is a typed, chainable workflow runner — define nodes, wire state through execute, and finalize outputs with returns. State types evolve as you add nodes, so the final output mapper is fully type-checked. Independent node executes are planned as a safe DAG optimization when their metadata reads and writes do not conflict.

import { ai, AxAIOpenAIModel, AxGEPA, flow } from "@ax-llm/ax";

const emailFlow = flow<{ emailText: string }>()
  .description("Email Priority", "Classify priority and write a one-line rationale.")
  .n("classifier", 'emailText:string -> priority:class "high, normal, low"')
  .n("rationale", "emailText:string, priority:string -> rationale:string")
  .e("classifier", (s) => ({ emailText: s.emailText }))
  .e("rationale", (s) => ({ emailText: s.emailText, priority: s.classifierResult.priority }))
  .r((s) => ({
    priority: s.classifierResult.priority,
    rationale: s.rationaleResult.rationale,
  }));

Tune the whole flow with GEPA (multi-objective Pareto optimizer). Define a metric that returns one or more named scores; GEPA explores the prompt space and returns a Pareto front.

const student = ai({ name: "openai", apiKey: process.env.OPENAI_APIKEY!,
  config: { model: AxAIOpenAIModel.GPT4OMini } });
const teacher = ai({ name: "openai", apiKey: process.env.OPENAI_APIKEY!,
  config: { model: AxAIOpenAIModel.GPT4O } });

const optimizer = new AxGEPA({
  studentAI: student,
  teacherAI: teacher,
  numTrials: 16,
  minibatch: true,
  minibatchSize: 6,
  seed: 42,
});

const result = await optimizer.compile(
  emailFlow,
  trainSet,
  async ({ prediction, example }) => ({
    accuracy: prediction.priority === example.priority ? 1 : 0,
    brevity: (prediction.rationale?.length ?? 0) <= 60 ? 1 : 0.4,
  }),
  { auto: "medium", validationExamples: valSet, maxMetricCalls: 240 },
);
// result.paretoFront, result.hypervolume, result.paretoFrontSize

The current published package is TypeScript / JavaScript:

npm install @ax-llm/ax

Generated Python, Java, C++, Go, and Rust libraries are checked in under packages/ and verified in this repo. They are prepared for ecosystem release as axllm, dev.axllm:ax, axllm::axllm, github.com/ax-llm/ax/go, and the Rust crate axllm. Until those registry lanes are enabled, use the repo runner to smoke-test the committed packages locally.

Optional packages:

npm install @ax-llm/ax-ai-aws-bedrock     # AWS Bedrock provider
npm install @ax-llm/ax-ai-sdk-provider    # Vercel AI SDK v5 integration
npm install @ax-llm/ax-tools              # MCP stdio transport, JS runtime extras

Get started

• Quick Start
• Runnable examples
• Multi-language release shape
• Compiler architecture
• DSPy concepts
• Signatures

Deep dives

• AI providers
• Audio I/O
• AxFlow workflows
• Optimization (GEPA, ACE)
• AxAgent & RLM
• Advanced RAG

OPENAI_APIKEY=your-key npm run tsx ./src/examples/<name>.ts
npm run example -- list
npm run example -- python axagent_pipeline.py
npm run example -- java AxFlowProgramGraphExample.java
npm run example -- cpp realtime_audio_events.cpp
npm run example -- go signature_schema.go
npm run example -- rust signature_schema.rs
npm run example -- ts src/examples/mcp-scripted-tools.ts
npm run example -- python mcp_scripted_tools.py
npm run example -- python axgen_openai_api.py
npm run example -- java AxGenOpenAIExample.java
npm run example -- cpp axgen_openai_api.cpp
npm run example -- go axgen_openai_api.go
npm run example -- rust axgen_openai_api.rs

npm run example -- list shows no-key and provider-api examples for TypeScript, Python, Java, C++, Go, and Rust. No-key examples cover signatures, AxAgent, AxFlow, MCP scripted transports, audio/realtime mapping, runtime adapters, optimizer artifacts, and GEPA with deterministic local clients. Provider API examples call real provider HTTP and read credentials from .env. TypeScript examples live under src/examples; generated language examples are canonical in packages/<language>/examples and are resolved from those packages first.

Highlights: extract.ts, react.ts, agent.ts, streaming1.ts, multi-modal.ts, audio-chat.ts, audio-batch-and-agent.ts, standard-schema.ts, rlm-memories-and-skills.ts, rlm-discovery.ts, gepa-flow.ts, openai-compatible.ts, ax-flow-enhanced-demo.ts. Browse all examples →

• Discord — questions and discussion
• Twitter — updates
• GitHub — source and issues
• DeepWiki — AI-generated docs

Ax is TypeScript-first. Most contributors and coding agents should focus on the TypeScript source change they are making and should not try to update every generated language backend by hand.

When a PR changes portable behavior under src/ax/ai/, src/ax/dsp/, src/ax/agent/, src/ax/flow/, or src/ax/mcp/, CI will ask for either AxIR/conformance updates or an AxIR backlog entry. If you are not already working in AxIR, use the backlog path:

npm run axir:backlog -- add --title "..." --surface axai --impact "..." --paths src/ax/ai/...
npm run axir:backlog:validate

That keeps normal TypeScript PRs small while giving AxIR maintainers and coding agents a precise queue for migrating the behavior into Python, Java, C++, Go, and future generated backends later.

• Author: @dosco
• GEPA and ACE optimizers: @monotykamary

Apache 2.0