[Mehedi Hasan (Mahdi-Hasan)]

Motivation & Context

When a mistralai.client.Mistral is injected, RawMistralChatClient, MistralChatClient and MistralEmbeddingClient still keep MISTRAL_SERVER_URL (or an explicit server_url=) in self.server_url. Requests go to the injected client's endpoint, but service_url() and telemetry report the other URL.

Description & Review Guide

• What are the major changes?
  • For an injected SDK client, server_url now always comes from client.sdk_configuration.get_server_details()[0].
  • Docstrings note that server_url is ignored when client is provided.
• What is the impact of these changes?
  • Only reported metadata changes. Requests already used the injected client's endpoint.
  • Owned-client and http_client paths keep their existing precedence. Model resolution is unchanged.
• What do you want reviewers to focus on?
  • The issue also suggests not loading API key / server URL settings at all on the injected path. load_settings reads every field of the TypedDict, and both fields are plain strings that can't fail parsing. So I kept one load_settings call and stopped using those values, rather than adding a second settings type.

Related Issue

Fixes #9965

AI Assistance

• No material AI assistance was used.
• This is an AI-assisted contribution. I reviewed, understood, and verified all submitted content and accept responsibility for it.

AI assistance details: implementation, tests, and this description. The issue's repro was re-run against a real Mistral SDK client before and after the change.

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR links to an agreed issue with no competing open PR, or the Related Issue section documents a trivial-change or repository-automation exception.
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[Copilot]

🟡 Changes recommended

Active moderate issues remain because injected SDK clients still load API key and server URL settings from ambient configuration, and the related documentation does not clearly distinguish SDK-client behavior from the deprecated HTTP-client path.

4 open issues

• Avoid loading API settings for an injected Mistral client · New
• Avoid loading API settings for an injected embedding client · New
• Clarify server_url behavior for injected SDK clients · New
• Clarify server_url behavior for injected SDK clients · New

_{Review effort: Lite}

[Copilot]

🟢 Approval recommended

No active review issues remain; the endpoint precedence and model-only settings behavior are covered by the added chat and embedding tests, with only non-blocking documentation and typing nits noted.

1 open issue

• Avoid loading API settings for an injected Mistral client

3 resolved since last review

• Avoid loading API settings for an injected embedding client
• Clarify server_url behavior for injected SDK clients
• Clarify server_url behavior for injected SDK clients

_{Review effort: Lite}

[Eduard van Valkenburg (eavanvalkenburg)]

Mehedi Hasan (@Mahdi-Hasan) The extra MistralChatModelSettings and MistralEmbeddingModelSettings types are not needed for this fix. required_fields controls validation after loading; it does not control which fields load_settings reads. Please keep the regular settings types, set the required fields for the injected-client path, and ignore the unused API-key and server-URL values. If we want field-selective settings loading, that should be added once to load_settings instead of duplicating settings schemas and adding type-ignore assignments here.

[Mehedi Hasan (Mahdi-Hasan)]

Eduard van Valkenburg (@eavanvalkenburg) Thanks, done in 2152a92.

• Removed MistralChatModelSettings and MistralEmbeddingModelSettings, along with the type: ignore assignments.
• Both clients make one load_settings call with the regular settings type. When a client is injected, the required fields are relaxed ([] for chat, ["embedding_model"] for embeddings). The loaded api_key and server_url are then ignored on the SDK-client path.
• Removed the two tests that spied on which settings type was loaded. The endpoint-precedence tests stay.

[Copilot]

🔵 Needs a closer look

The injected-SDK paths still validate credential and server settings that are ignored afterward, allowing invalid overrides to break construction; both clients need model-only schema resolution and regression coverage.

No open issues 🎉

1 resolved since last review

• Avoid loading API settings for an injected Mistral client

_{Review effort: Lite}

[Mehedi Hasan (Mahdi-Hasan)]

Eduard van Valkenburg (@eavanvalkenburg) The python-tests-coverage failure (and merge-gatekeeper with it) is not from this change. It is a test-order flake in core.

• Failing test: test_function_invocation_logic.py::test_mixed_declaration_only_batch_defers_siblings_until_all_host_results, with KeyError: 'value' in _sessions.py:437.
• Cause: test_sessions.py::test_registered_child_tag_does_not_hijack_message_payload registers a global "text" state type and never unregisters it. When both tests run on the same xdist worker, the leaked decoder catches the {"type": "text", ...} content items during session restore.
• This PR only touches packages/mistral. All 9 Python Tests matrix jobs pass.

I can't re-run the job from a fork. Could you re-run it when you review?