[poissoncorp]

AI conversations can now override the output schema for one turn with a schema string or a sample object, get plain text answers with run_text() and stream_text(), and register tools that take no parameters.

• Changed: run_with_schema and stream_with_schema accept an AiOutputOptions, a JSON schema string or a sample object (dict or entity). New run_text(), stream_text(on_chunk), handle_no_args() and receive_no_args() on AiConversation. RunConversationCommand streams whenever stream_property_path is not None, so a no-schema turn can stream with an empty path, as in C#.
• Already present: AiOutputOptions, OutputOptions in the request body and string Response parsing.
• Not ported: checks of the options against the answer type, because Python answers are untyped (the no_schema and schema conflict is already rejected by AiOutputOptions). Sync and async overload pairs and expression-based paths are C#-specific.
• Tests: test_ai_output_options.py gains schema shorthand, raw text and no-args tool tests with mocked conversation calls, adapted from RavenDB_24824. The RavenDB_24824 tests that need a live OpenAI connection are not ported.
• Checks: black --check . passed; test_ai_output_options passed (31); ai_agent_tests passed, and the tests that need a server run only in CI.

Decision needed

Python has no overloading, and run() already means "use the agent's schema", so the C# overloads need names.

1. Separate methods run_text(), stream_text(), handle_no_args(), receive_no_args() (recommended, this PR): explicit, and existing signatures stay the same.
2. An optional output_options argument on run() and stream(), as in the Node.js client.
3. handle() and receive() check the handler's parameter count instead of new methods.

Existing behavior that this PR does not change: add_action_response keeps only string results, so a handler that returns a dict sends no content, while C# serializes it to JSON.

---

Ticket: RDBC-1151
C# change: ravendb/ravendb#22632

SDK Sync Agent

[poissoncorp]

The failing check is not caused by this PR. No code change.

• Two tests fail, both in test_secured_subscription_basic: test_should_send_all_new_and_modified_docs and test_should_stream_all_documents_after_subscription_creation. Both time out waiting for the first batch (queue.Empty).
• The same two tests failed at the same time in five open PRs with unrelated changes (RDBC-1148 Merge release/quill into ravendb/v7.2 #313, RDBC-1141 RavenDB-25401 - Fix refusal on streaming #314, RDBC-1140 RavenDB-27181 : GPT-5.6 models unusable with AI agents - tools rejected on /v1/chat/completions #315, RDBC-1143 RavenDB-27175 Subscriptions - make it easier to inspect the current state #317, RDBC-1151 RavenDB-24824 Add per-call output schema override and string output for AI conversations #318). Each time they failed on only one or two Python versions and passed on the others with the same code.
• RDBC-1146 RavenDB-27173 Advanced.RegisterForConcurrencyCheck #312 passed on all Python versions earlier the same day.

A rerun of the failed job should pass. If it fails again, the secured test server setup in CI needs a look.

SDK Sync Agent

[poissoncorp]

please fix failing/flaky test as well and rebase other PRs after merge

[poissoncorp]

Fixed the flaky secured subscription tests in 19fd21f.

• Cause: test_secured_subscription_basic waited 4 seconds for each batch. The worker waits 5 seconds (time_to_wait_before_connection_retry) before it reconnects, so one failed first connection attempt made Queue.get time out.
• Fix: the wait is now 15 seconds, the same as the other subscription tests in the suite.
• The failing build (3.12, 7.2) job failed only on these two tests. The failures on RDBC-1148 Merge release/quill into ravendb/v7.2 #313, RDBC-1141 RavenDB-25401 - Fix refusal on streaming #314, RDBC-1140 RavenDB-27181 : GPT-5.6 models unusable with AI agents - tools rejected on /v1/chat/completions #315, RDBC-1156 RavenDB-22083 Report a proper IndexCompilationException for a C# index whose outer from clause does not enumerate the documents source #316 and RDBC-1143 RavenDB-27175 Subscriptions - make it easier to inspect the current state #317 are the same two tests.
• Rebase: the other PRs will be updated from v7.2 after this PR is merged.

SDK Sync Agent

[poissoncorp]

fix failing test

[poissoncorp]

Fixed the failing test_more_like_this in 83f176c.

• Cause: posts/3 and posts/4 each share one term with posts/1, and both terms occur in three documents, so the two posts score almost the same. The latest 7.2 nightly server returns them in the other order. The test asserted a fixed order.
• Fix: posts/2 must still be the first result; the other two results may come in either order.
• The same failure is on all Python versions of RDBC-1162 RavenDB-27614 Generate and validate identifiers for server-wide AI connection strings #319 and RDBC-1161 RavenDB-27289 Require a value for every declared agent parameter #320, which do not touch indexing or queries. Those PRs pick up the fix when they are updated from v7.2 after this PR is merged.
• The Java client asserts the same fixed order. If the new order is a server regression rather than a scoring change, the server team has to look at it; the test no longer reports it.
• The test needs a server and ran only in CI. black --check passed.

SDK Sync Agent