Ax Examples Guide

This page lists every runnable example in src/examples/, grouped by theme and linked to GitHub for easy browsing.

Getting Started

• chat.ts - Basic chat example with function calling and conversation flow.
• extract.ts - Extract structured fields from free-form text.
• extract-test.ts - Small extraction test case with typed outputs.
• summarize.ts - Summarize a list of updates into multiple output formats.
• simple-classify.ts - Lightweight text classification with labeled classes.
• marketing.ts - Generate marketing-style copy from prompts.
• embed.ts - Create embeddings and inspect the returned vectors.
• prime.ts - Simple reasoning task around prime number generation.
• fibonacci.ts - Use an AI call plus runtime execution to compute Fibonacci numbers.

Signatures, Structured Output, and Examples

• structured_output.ts - Structured output with validation constraints and nested objects.
• fluent-signature-example.ts - Build signatures with the fluent f() API.
• debug_schema.ts - Inspect generated JSON schema from constrained signatures.
• use-examples.ts - Attach few-shot examples to a generator.
• sample-count.ts - Generate multiple samples from a single prompt program.
• result-picker.ts - Pick the best result from multiple generated samples.

Functions and Tool Calling

• function.ts - Basic function-calling setup with Ax.
• signature-tool-calling.ts - Combine agent signatures with tool execution.
• function-result-formatter.ts - Customize how tool results are formatted back to the model.
• function-result-picker.ts - Choose among multiple tool results before returning an answer.
• stop-function.ts - Stop a tool-driven workflow with a terminating function.
• food-search.ts - Tool-driven restaurant search and selection flow.
• react.ts - ReAct-style prompting with tools and intermediate reasoning.
• smart-home.ts - Multi-tool smart home assistant simulation.

Streaming, Assertions, and Control Flow

• streaming.ts - Stream structured results while validating output fields.
• streaming-asserts.ts - Enforce line-by-line validation during streaming.
• asserts.ts - Add output assertions and retry behavior.
• abort-simple.ts - Cancel a request with AbortController.
• abort-patterns.ts - Compare several request cancellation patterns and edge cases.

Agents and Agent Patterns

• agent.ts - Basic multi-agent composition with a shared runtime.
• agent-migration-example.ts - Migrate to the newer agent() factory API.
• customer-support.ts - Parse support emails into structured support fields.
• meetings.ts - Meeting-oriented assistant workflow example.
• show-thoughts.ts - Surface provider reasoning or thought traces when supported.
• self-improving-agent.ts - Persist traces and checkpoints for a self-improving workflow.
• codingWithMemory.ts - Agentic coding flow backed by an MCP memory server.

RLM and Runtime-State Examples

• rlm.ts - Core RLM example for long-context analysis with runtime tools.
• rlm-test.ts - Compact RLM test agent with namespaced tools.
• rlm-discovery.ts - Runtime discovery of tools and permissions in an agent.
• rlm-long-task.ts - Context-policy example for long-running tasks and checkpoint summaries.
• rlm-truncated-context.ts - Keep only the most relevant tail of a long conversation in prompt context.
• rlm-shared-fields.ts - Propagate shared fields automatically into subagents.
• rlm-adaptive-replay.ts - Collapse older turns into checkpoint summaries during replay.
• rlm-live-runtime-state.ts - Inspect the structured live runtime-state block in prompts.
• rlm-clarification-resume.ts - Pause for clarification, save state, and resume later.
• rlm-agent-controlled.ts - Agent-controlled workflow termination and clarification handling.
• rlm-agent-optimize.ts - Optimize an RLM agent and persist the resulting artifact.

Flow Orchestration

• ax-flow.ts - End-to-end AxFlow example with nodes, descriptions, and execution.
• ax-flow-enhanced-demo.ts - Explore richer AxFlow composition patterns.
• ax-flow-async-map.ts - Use async mapping steps inside a flow.
• ax-flow-auto-parallel.ts - Demonstrate AxFlow automatic parallelization.
• ax-flow-map-merge-test.ts - Test mapping and merge behavior in flows.
• ax-flow-signature-inference.ts - Infer flow signatures from connected nodes.
• ax-flow-to-function.ts - Convert flow logic into function-like execution patterns.
• fluent-flow-example.ts - Build flows with the fluent builder API.
• flow-type-inference-demo.ts - Show state evolution and type inference across flow steps.
• flow-type-safe-output.ts - Produce strongly typed final outputs from a flow.
• flow-logging-simple.ts - Attach a simple color logger to a flow.
• flow-verbose-logging.ts - Enable more verbose runtime logging for flow execution.

Optimization, Training, and Evaluation

• teacher-student-optimization.ts - MiPRO teacher-student optimization with a small model target.
• mipro-python-optimizer.ts - Run MiPRO with the Python optimizer backend.
• simple-optimizer-test.ts - Minimal optimizer setup for a classification task.
• optimizer-metrics.ts - Track and inspect optimizer metrics over time.
• checkpoint-recovery.ts - Save and restore optimization checkpoints.
• gepa.ts - Introductory GEPA optimization example.
• gepa-flow.ts - Apply GEPA optimization to a flow-based program.
• gepa-train-inference.ts - Train with GEPA and reuse the result for inference.
• gepa-quality-vs-speed-optimization.ts - Multi-objective GEPA example balancing quality and speed.
• ace-train-inference.ts - Train and update an ACE playbook from labeled examples.

RAG, Retrieval, and Memory

• vectordb.ts - Insert text into an in-memory vector DB and query it.
• rag-docs.ts - Build a simple document ingestion and retrieval pipeline.
• advanced-rag.ts - Extend RAG with custom retrieval behavior.

Multi-Modal, Audio, and Vision

• multi-modal.ts - Basic multimodal prompt with image input.
• multi-modal-abstraction.ts - Higher-level abstractions for multimodal workflows.
• image-arrays-test.ts - Send multiple images in a single request.
• image-arrays-multi-provider-test.ts - Compare multi-image support across providers.
• audio-arrays-test.ts - Send multiple audio clips and inspect the response.

Provider-Specific Examples

Anthropic

• anthropic-thinking-function.ts - Combine Claude thinking mode with function calls.
• anthropic-thinking-separation.ts - Capture reasoning separately from final output.
• anthropic-web-search.ts - Use Anthropic web search tooling through Ax.
• test-anthropic-cache.ts - Exercise Anthropic prompt caching behavior.

Gemini and Vertex

• gemini-context-cache.ts - Reuse cached context with Gemini models.
• gemini-empty-params-function.ts - Call Gemini tools with an empty parameter schema.
• gemini-file-support.ts - Upload files and reference them in Gemini prompts.
• gemini-function-cache.ts - Combine Gemini function calls with caching.
• gemini-google-maps.ts - Connect Gemini with Google Maps tool usage.
• gemini-parallel-test.ts - Try parallel Gemini requests and compare outputs.
• vertex-auth-example.ts - Authenticate against Vertex AI with dynamic Google auth.

OpenAI

• openai-responses.ts - Use the OpenAI Responses API through Ax.
• openai-web-search.ts - Run a web-search-enabled OpenAI example.
• reasoning-o3-example.ts - Reasoning-focused example using OpenAI o3.

Other Providers

• grok-live-search.ts - Query Grok with live search enabled.
• openrouter.ts - Route requests through OpenRouter.

MCP Integrations

• mcp-client-memory.ts - Connect to an MCP memory server.
• mcp-client-blender.ts - Drive Blender through an MCP client.
• mcp-client-pipedream.ts - Use Pipedream via MCP transport.
• mcp-client-notion-http-oauth.ts - Connect to Notion MCP over HTTP with OAuth.
• mcp-client-notion-sse-oauth.ts - Connect to Notion MCP over SSE with OAuth.

Routing, Balancing, and Service Selection

• balancer.ts - Balance requests across multiple configured services.
• ax-multiservice-router.ts - Route requests between providers with shared model aliases.

Observability and Debugging

• debug-logging.ts - Enable debug logging for model calls and tool steps.
• telemetry.ts - Export traces with OpenTelemetry.
• metrics-export.ts - Export runtime metrics for external monitoring.

Browser, Local Runtime, and Deployment

• web-chat.html - Browser chat UI using the Ax bundle.
• webllm-chat.html - Browser chat example powered by WebLLM.
• cors-proxy.js - Small CORS proxy helper for browser-based development.
• docker.ts - Run tool-like tasks inside a Docker session.