Examples

Detailed practical examples for common NucleusIQ workflows.

Updated for v0.7.14: Autonomous harness (termination_reason, diagnostics, Task.resources, result.parsed). See migration notes if upgrading.

Getting started

• Basic agent — Your first agent with tools (OpenAI, Gemini, context management)
• Streaming — Real-time token-by-token output

Workflows

• File workflow — Search, read, and extract from files
• Autonomous mode guide — why the harness changed, what improved, and current APIs
• Autonomous workflow — Invoice extraction with Task.resources, enforced schema, and diagnostics (v0.7.14)

Observability

• Usage tracking — Token usage by purpose and origin
• Cost estimation — Dollar cost tracking after execution

Provider examples

• OpenAI-compatible quickstart — Self-hosted vLLM / SGLang / llama.cpp / LM Studio / Azure OpenAI v1 (nucleusiq-openai-compatible 0.1.0, nucleusiq>=0.7.13)
• Gemini quickstart — Google Gemini with all three execution modes
• Anthropic quickstart — Claude DIRECT / STANDARD / AUTONOMOUS (nucleusiq-anthropic 0.2.2, nucleusiq>=0.7.12)
• Groq quickstart — Groq DIRECT / STANDARD / AUTONOMOUS (nucleusiq-groq 0.1.1, nucleusiq>=0.7.12)
• Ollama quickstart — Native Ollama DIRECT / STANDARD (nucleusiq-ollama 0.2.2, nucleusiq>=0.7.12)
• OpenAI-compatible provider — Engines, auth, validate(), structured-output policy
• Anthropic provider guide — Phase B native tools, prompt caching, extended thinking
• Groq provider guide — 429 / Retry-After, strict_model_capabilities, repo scripts
• Ollama provider guide — Native /api/chat, think, vision, structured-output + tools caveat

Tool adapter examples

• MCP quickstart — Universal Model Context Protocol adapter across every provider (nucleusiq-mcp 0.1.1, nucleusiq>=0.7.12): stdio, HTTP + Bearer, multi-server, filter/rename, graceful degradation, source tracing, plugin guardrails
• MCP integration guide — Transports, auth strategies, filtering, ping(), comparison vs OpenAI server-side MCP

Which page for self-hosted / BYOM?

Start with OpenAI-compatible quickstart; use OpenAI-compatible provider for engines, auth, validate(), and Azure OpenAI v1 limits.

Which page for Anthropic (Claude)?

Start with Anthropic quickstart; use Anthropic provider for install pins, env vars, limitations, and CI-tested examples.

Which page for Ollama?

Start with Ollama quickstart; use Ollama provider for the full capability matrix and env reference.

Which Groq page?

Start with Groq quickstart for copy-paste snippets; use Groq provider for deeper operational guidance.

Which page for MCP?

Start with MCP quickstart for copy-paste recipes (stdio + HTTP, multi-server, tracing); use MCP integration guide for transports, OAuth, decorator filters, and graceful degradation.

Repository examples

Full runnable scripts are available in the GitHub repository:

• Core examples
• OpenAI examples
• OpenAI-compatible examples
• Gemini examples
• Anthropic examples — DIRECT through AUTONOMOUS, streaming, native structured demo
• Groq examples — Direct through Autonomous + structured output
• Ollama examples — Smoke, Direct, streaming live, capabilities matrix
• MCP examples — Eight runnable examples: stdio, HTTP + auth, multi-server, OAuth, error handling, health check, decorator filters, full agent with LLM
• Notebooks — Jupyter notebooks: context management showcase, mcp_tools_showcase.ipynb (Windows-friendly Streamable HTTP demo), and more