Ollixir Examples

Examples demonstrating Ollixir features, designed to run against a real Ollama server.

Run everything

./examples/run_all.sh

The runner will:

• Verify the Ollama server is reachable (default: http://localhost:11434).
• Pull required models if the ollama CLI is available (llama3.2, nomic-embed-text, llava, deepseek-r1:1.5b, codellama:7b-code).
• Run every example (except the MCP stdio server), including the interactive chat history and LiveView module compile.
• Skip cloud/web examples if OLLAMA_API_KEY is not set (invalid keys report 401/403).
• Multimodal examples require a compatible model (llava). If not installed, they print a prompt.
• Thinking examples use deepseek-r1:1.5b (supports think).
• Skip custom model creation unless RUN_CREATE_MODEL=1 (it writes a new local model).

Run ./examples/run_all.sh --help to see optional skips.

Run a single example

elixir examples/basic/chat.exs

Example map

Basic operations

• chat.exs - Simple chat request
• completion.exs - Text completion
• list_models.exs - List available models
• show_model.exs - Model details

Streaming

• chat_stream.exs - Stream to enumerable
• chat_stream_pid.exs - Stream to process (PID)
• completion_stream.exs - Stream completion output
• liveview_chat.ex - Phoenix LiveView module example (prints snippet if LiveView isn't available)

Conversations

• chat_history.exs - Multi-turn chat (type quit to exit)
• context_continuation.exs - Continue using context
• system_prompts.exs - System role messages

Tools / Function Calling

• basic_tools.exs - Simple tool use
• multi_tool.exs - Multiple tools
• function_tools.exs - Function → tool conversion
• tool_loop.exs - Agentic tool loop

Structured Output

• json_format.exs - format: "json"
• json_schema.exs - Schema enforcement
• with_ecto.exs - Validate with Ecto changeset

Multimodal

• image_chat.exs - Image + chat
• image_completion.exs - Image + completion Sample image: assets/ollixir.png (PNG/JPEG recommended for maximum compatibility).

Thinking Mode

• basic_thinking.exs - Enable thinking
• thinking_stream.exs - Stream thinking + response

Embeddings

• single_embed.exs - Single embedding
• batch_embed.exs - Batch embeddings
• similarity.exs - Cosine similarity

Model Management

• pull_progress.exs - Pull with progress
• create_custom.exs - Create a custom model
• preload_unload.exs - Load/unload models

Web (Cloud API)

• web_search.exs - Web search (requires API key)
• web_fetch.exs - Web fetch (requires API key)

MCP (Model Context Protocol)

• mcp_server.exs - MCP stdio server for web_search/web_fetch (requires API key) Works with any MCP client that supports stdio (Cursor, Claude Desktop, Cline, Continue, Open WebUI).

HuggingFace Integration

• hf_gguf_integration.exs - End-to-end demo: discover GGUF files via hf_hub, auto-select quantization, pull from HuggingFace, and run inference (~0.1GB SmolLM2-135M)
• hf_live_test.exs - Pull and chat with HF models (downloads ~0.7GB Llama-3.2-1B)
• Uses Ollixir.HuggingFace module for model discovery and auto-selection
• Requires {:hf_hub, "~> 0.1.3"} - See HuggingFace Integration Guide

Advanced Patterns

• chat_logprobs.exs - Chat log probabilities
• completion_logprobs.exs - Completion log probabilities
• fill_in_middle.exs - Fill-in-middle completion
• concurrent_requests.exs - Task.async_stream
• rate_limiting.exs - Simple throttling
• error_handling.exs - Request/response errors
• options_presets.exs - Options presets
• typed_responses.exs - Response structs
• genserver_integration.exs - GenServer wrapper

Requirements

• Elixir 1.15+
• Ollama installed and running (ollama serve)
• Models pulled (ollama pull llama3.2, ollama pull nomic-embed-text, ollama pull llava, ollama pull deepseek-r1:1.5b, ollama pull codellama:7b-code)
• For web examples: create a key at https://ollama.com/settings/keys and set OLLAMA_API_KEY

To pull everything at once:

./examples/install_models.sh