A stream_options.include_usage usage chunk (empty delta plus usage) was folded into the final
transform round and rebuilt without its usage, so token counts and cost vanished from clients.
Metadata-only chunks are now replayed after the final text flush.
A terminal tool-call chunk arriving while earlier text was still held back carried
finish_reason=tool_calls ahead of that text. The finish_reason is now deferred to the final
text chunk whenever the choice has held text.
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Forward streaming_transform_mode from guardrail litellm_params into PromptSecurityGuardrail so incremental_diff is reachable from config; the default stays block_only. In incremental_diff the guardrail now returns stream_holdback_chars alongside the rewritten texts so that a value split across streamed chunks (or across an abbreviation period) is never partially released before the vendor rewrite arrives. Each response text gets its own protect call so modified_text maps back to the right choice when n > 1, and custom_guardrail no longer logs a clean response as mask just because the guardrail attached holdback metadata
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Trimming the streamed buffer to the retained tail could drop a category
exception phrase that suppresses a later keyword, or the identifier word
of an unfinished sentence that a conditional category pairs with a later
block word. Refuse the cut while either would leave the buffer so the
bounded scan masks and blocks exactly like a scan of the full text
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
The streaming post-call hook rescanned the whole accumulated choice buffer on every chunk, so scan cost grew quadratically with output length. Keep a bounded per-choice buffer instead: once it exceeds twice the scan context, drop the head when masking the head and tail separately yields the same output as masking the whole buffer, so no pattern, phrase or exception straddles the cut. Detections from the dropped head are kept and merged, deduplicated, into the final log row
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Removes the Singulr async_logging_hook and logging_hook overrides so logging_only runs through CustomGuardrail.async_logging_hook: the response scope reaches Singulr as an assistant message instead of a raw ModelResponse dump, a vendor timeout is recorded as guardrail_failed_to_respond, a request-scope block ends the scan, and the sync success callback thread makes no Singulr call.
Decides MCP versus LLM by the proxy logging object's call_type (then the call_type or server-only markers in request_data), never by name, arguments or mcp_tool_name keys a client can put in a chat body. REST /mcp-rest/tools/call pre-scans reach Singulr as mcp_request and a non-mapping arguments value is forwarded as tool_arguments instead of raising.
should_block is a strict bool defaulting to false so a null verdict is an invalid response that block_on_error decides; payload fields drop Any for Sequence, Mapping and AssistantMessage types; metadata carries only the keys present; docstrings and section comments removed per the repo comment policy.
Squash of BerriAI/litellm#37464 (head da298ca7) by @aniket-kardile, adopted onto main: v2 gateway payload contract with request, response, mcp_request and mcp_response scopes, typed payload models, proxy user, org and team metadata forwarded to Singulr, and the logging_only, pre_mcp_call and post_mcp_call modes.
On pre_call, during_call and post_call the hook data is the client request body, so a
client-supplied litellm_params.metadata.internal_call_origin must not skip enforcement.
Type the during_call helper's request payload as dict[str, object]
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
The request-side prompt told the judge to focus on the most recent user turn but the text under review was every extracted request message joined together, so a multi-turn request with an off-topic earlier turn and an on-topic latest turn scored 50 and was blocked. Request-side judging now evaluates the last extracted request text (after the configured message scoping) and passes the full role-labelled conversation only as context. Response-side judging still evaluates all extracted response text
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Judge sub-calls now carry the internal_call_origin metadata stamp and the
guardrail skips any logged call bearing it, so a logging_only judge no longer
recurses into an unbounded chain of judge requests
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Response-side guardrail scans on OpenAI Chat Completions, Anthropic Messages, and OpenAI Responses now carry structured_messages (the request turns scoped exactly like the pre-call scan, closed by the model's reply as an assistant turn) and tools (the request's function definitions), in addition to texts, images, and tool_calls.
Guardrails that used structured_messages or tools as a response-side signal (akto, crowdstrike_aidr, hiddenlayer, openai moderations, promptguard, qualifire, straiker) keep their previous response payloads. Logging-only scans whose output translation differs from the input translation get a chat-shaped request so the context survives.
Resolves LIT-6628
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
