feat: add OVHcloud AI Endpoints built-in provider

[aheritier]

Summary

Add ovhcloud as a first-class built-in alias provider for OVHcloud AI Endpoints, following the same pattern as the recently added baseten provider (PR #3341).

Motivation

Users are already using OVHcloud AI Endpoints today, but they must write verbose custom provider config (as shown in #3145):

providers:
  ovhcloud:
    api_type: openai_chatcompletions
    base_url: https://oai.endpoints.kepler.ai.cloud.ovh.net/v1
    token_key: OVH_AI_ENDPOINTS_ACCESS_TOKEN

models:
  ovh-qwen:
    provider: ovhcloud
    model: Qwen3.5-397B-A17B

With a built-in alias, users write just:

agents:
  root:
    model: ovhcloud/Qwen3.5-397B-A17B

OVHcloud AI Endpoints is noteworthy because several models (including Qwen3.5-397B-A17B) offer a free tier with no API key required (up to 2 req/s), making it an excellent option for trying docker-agent without any credentials. opencode.ai already supports this provider under the same alias name.

Implementation Plan

This is a straightforward alias provider — no new API client or streaming adapter needed. The streaming delta.reasoning quirk from #3145 is already handled in main.

Changes required

1. Register the built-in alias in the provider registry (Go code) with:

   • Alias: ovhcloud
   • API type: openai_chatcompletions
   • Base URL: https://oai.endpoints.kepler.ai.cloud.ovh.net/v1
   • Token env var: OVH_AI_ENDPOINTS_ACCESS_TOKEN

2. Documentation — new page docs/providers/ovhcloud/index.md covering setup, usage (inline + named model), available models, and a code-assistant example

3. Provider overview — add ovhcloud row to both tables in docs/providers/overview/index.md

4. Example YAML — examples/ovhcloud/agent.yaml

Notable models to document

Model	Notes
Qwen3.5-397B-A17B	Free tier (no API key), 2 req/s
Qwen2.5-72B-Instruct	Qwen 2.5 family
Llama-3.3-70B-Instruct	Meta Llama 3.3
Mistral-7B-Instruct-v0.3	Mistral 7B
DeepSeek-R1	DeepSeek reasoning

References

• OVHcloud AI Endpoints overview: https://endpoints.ai.cloud.ovh.net/
• OVHcloud docs: https://docs.ovhcloud.com/fr/guides/public-cloud/ai-machine-learning/ai-endpoints/overview
• opencode.ai provider reference: https://opencode.ai/docs/providers/#ovhcloud-ai-endpoints
• Related issue: Stream silently produces empty response when model outputs non-standard delta.reasoning SSE field (Qwen3 thinking mode) #3145
• Baseten provider PR (reference implementation): feat: add Baseten provider support #3341

[aheritier]

not sure about the default model @Sayt-0 @philippart-s

[Sayt-0]

I will check

[philippart-s]

not sure about the default model @Sayt-0 @philippart-s

We have no default model but you can take oss-gpt-120b as default model as it's our most used model 😊.

We have also an endpoint to list all available models if you want.

[Sayt-0]

In the firtst commit of the PR the models documented were false but now it's fixed and merge based on the OVH documentation !

[Sayt-0]

Let me know if we have to adjust the doc

[philippart-s]

Let me know if we have to adjust the doc

Once it's available, I'll check 😉