fix: use Ollama OpenAI-compatible endpoint (/v1) by default

[justi]

Problem

The Ollama provider inherits from OpenAI but uses the native API base (e.g. http://localhost:11434) which doesn't serve OpenAI-compatible endpoints. This causes:

1. Model listing fails with 404 — provider tries GET /models but Ollama serves models at /api/tags (native) or /v1/models (OpenAI-compatible)
2. with_schema silently doesn't work — since models aren't in the registry, they don't have structured_output capability, so response_format is never sent

Steps to reproduce

RubyLLM.configure { |c| c.ollama_api_base = "http://localhost:11434" }
RubyLLM.models.refresh!
# => WARN: Failed to fetch Ollama models (RubyLLM::Error: 404 page not found)

chat = RubyLLM.chat(model: "gemma:latest")
# => RubyLLM::ModelNotFoundError: Unknown model: gemma:latest

Solution

Automatically append /v1 to ollama_api_base so the provider uses Ollama's OpenAI-compatible endpoint. Idempotent — does not double-append if /v1 is already present (as in the existing test configuration).

Since Ollama < OpenAI, using the /v1 endpoint means all OpenAI provider logic (chat, models, schemas) works without any other changes.

# Before: http://localhost:11434       → GET /models       → 404
# After:  http://localhost:11434/v1    → GET /v1/models    → 200 ✓

What works after this fix

• RubyLLM.models.refresh! fetches Ollama models correctly
• RubyLLM.chat(model: "gemma:latest") finds the model
• chat.with_schema(MySchema) sends response_format and returns parsed Hash

Edge cases handled

Input	Result
http://localhost:11434	http://localhost:11434/v1
http://localhost:11434/	http://localhost:11434/v1
http://localhost:11434/v1	http://localhost:11434/v1 (no change)
http://localhost:11434/v1/	http://localhost:11434/v1
https://my-ollama.com:8080	https://my-ollama.com:8080/v1

Note on native API

Ollama's native /api/chat endpoint with format param provides better nested JSON Schema enforcement than the OpenAI-compatible endpoint. A future enhancement could use the native endpoint for structured output while keeping /v1 for everything else.

Tests

5 new test cases for #api_base added to ollama_spec.rb, following existing spec style (subject, let, context, instance_double).

Tested locally with Ollama 0.17.5, gemma:latest (9B), llama3.2:3b (3B).

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR fixes the Ollama provider’s default API base so it targets Ollama’s OpenAI-compatible endpoints, which enables OpenAI-inherited behavior (model listing, chat, and structured output) to work correctly against a standard Ollama host URL.

Changes:

• Normalize ollama_api_base to always use the /v1 OpenAI-compatible endpoint (without double-appending and with trailing-slash cleanup).
• Update Ollama provider specs to cover #api_base normalization cases and to use a local config double (consistent with other provider specs).

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
lib/ruby_llm/providers/ollama.rb	Adjusts #api_base to normalize to the /v1 OpenAI-compatible base URL for Ollama.
spec/ruby_llm/providers/ollama_spec.rb	Adds test coverage for #api_base normalization and refactors header tests to use an isolated config double.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.