Responses

Configurations in Responses.

Configurations

• Agentic Loop Deferred Mcp Fixture — Minimal agentic-loop pipeline that sanitizes deferred MCP connectors on the first inference round. defer_loading: true skips tools/list, so replay never opens an independent MCP callout
• Agentic Loop Fixture — Minimal agentic loop pipeline for inference fixture replay
• Agentic Loop — Demonstrates the openai_agentic_loop filter with iterative_request_router for step-based model-tool-model looping in the Responses API
• Body Size Limits — Demonstrates how raw request body size is enforced across a chain of OpenAI Responses filters that each buffer the request body
• Client Tool Compat Chat Completions — Lets a rich Codex-style Responses client (POST /v1/responses with custom, namespace, local shell, and client-executed tool_search tools) reach a function-only Chat Completions backend (POST /v1/chat/completions) by composing openai_client_tool_compat with responses_to_chat_completions in one iterative-router step (GitHub issue #1206)
• Client Tool Compat — Lets a rich Codex-style Responses client talk to a function-only Responses backend (for example vLLM at POST /v1/responses) without routing through /v1/chat/completions and without executing client-owned tools in Praxis
• Codex Http Chat Translation — Release-acceptance configuration for GitHub issue #870. A Codex client speaks the OpenAI Responses API over HTTP while Praxis selects a Chat Completions-only provider, translates both request and streaming response, rewrites the provider path, and replaces the client credential
• Compact — Demonstrates compaction after rehydrate, file resolve, and document extract so rewritten current-turn content survives history replacement
• Doc Extract — Converts input_file content parts to input_text for inference backends that do not natively support input_file (e.g. vLLM, llm-d)
• File Resolve — Resolves file_id and file_url references in Responses API input by fetching file metadata and content, then inlining base64 content as file_data or image_url before forwarding
• File Search Callout — Demonstrates hosted file_search execution under the unified agentic loop (#1046). openai_agentic_loop is the sole loop owner: it parses each model response, records the file_search_call items it sees as assignments, and publishes the single continuation signal (action=loop|done). openai_file_search_callout is a pure request-phase dispatcher: at request- body EOS on each IRR re-entry it executes the assigned calls against the vector store and reconciles each item in place, then the owner prepares the next inference request
• File Search Chat Completions Fixture — Single-upstream fixture configuration for recording the private Chat Completions function representation of a Responses file_search tool
• File Search Chat Completions — Accepts finite OpenAI Responses requests with hosted file search while targeting a backend that only implements /v1/chat/completions
• File Search Streaming — Demonstrates streaming hosted file_search through the iterative_request_router
• Format Routing — Routes AI API traffic by request-head operation identity and body format
• Full Flow Agentic — Runs the complete Responses API pipeline through an agentic iterative_request_router that executes hosted file_search, web_search, and MCP tool calls in a model-tool-model loop, persisting both buffered and streaming (stream: true) responses
• Http Passthrough — The strict-HTTP acceptance test for pinned Codex CLI for GitHub issue #870 drives the proxy over POST /v1/responses and any other Responses API paths the client may probe
• Irr Terminal Streaming — Demonstrates a single-step iterative_request_router pipeline that exposes a native OpenAI Responses SSE body incrementally. openai_responses_proxy always advertises the streaming capability and selects Praxis’s typed streaming transport automatically for an effective “stream”: true request; there is no operator opt-in
• Mcp Dispatch — Demonstrates the openai_mcp_dispatch filter configuration
• Mcp Outbound Chain — Demonstrates binding an operator outbound_chain onto the outbound MCP callout made by openai_mcp_tool_resolve (the tools/list discovery request)
• Mcp Streaming — Demonstrates MCP tool calls over the filtered-subrequest transport with SSE streaming support
• Mcp Tool Resolve — Demonstrates the openai_mcp_tool_resolve filter, which resolves MCP tool entries in the Responses API tools array into concrete tool definitions by calling tools/list on each upstream MCP server
• Model Rewrite — Rewrites or injects the top-level model field in Responses API and Chat Completions request bodies before forwarding to the inference backend
• Rehydrate Fixture — Minimal native OpenAI Responses pipeline that stores a first turn, rehydrates a stored previous_response_id into the outbound input history, and proxies to a native /v1/responses backend
• Rehydrate — Validates previous_response_id by fetching the stored response, confirming its status is completed, and promoting the ID to filter metadata
• Request Validate — Validates Responses API JSON and enriches request metadata
• Response Store Postgres Mtls — Persists non-streaming Responses API responses to PostgreSQL over a TLS-verified connection that authenticates with a client certificate instead of a password
• Response Store — Persists non-streaming Responses API responses to a database and serves stored data via GET endpoints and handles DELETE /v1/responses/{id} locally
• Responses Proxy — Forward Responses API traffic through a Praxis AI listener to a compatible backend.
• Responses Routing — Routes Responses API traffic by detected mode
• Responses To Chat Completions Reasoning — Same pipeline as responses-to-chat-completions.yaml, but targets a vLLM backend that returns raw reasoning in choices[].message.reasoning
• Responses To Chat Completions — Translate the Responses API shape for an OpenAI-compatible Chat Completions backend.
• State Ownership — Provider-neutral owner isolation for persisted OpenAI Responses and Conversations
• Stream Events — Demonstrates the openai_stream_events filter, which composes the current iterative-request-router (IRR) execution into one logical Responses SSE stream: it parses each backend SSE chunk, accumulates state (response object, output items, tool calls, usage) into ResponsesState, normalizes the per-round lifecycle, and preserves parser state through stream completion
• Tool Routing — Demonstrates using openai_tool_parse to route Responses API requests by their tool composition
• Vector Stores Routing — Routes /v1/vector_stores traffic and all its subresources to a dedicated backend (any server compatible with the OpenAI Files / Vector Stores API), while sending everything else to a default backend
• Vllm Agentic Api — vLLM Agentic API: https://github.com/vllm-project/agentic-api
• Web Search Chat Completions Fixture — Single-upstream fixture configuration for recording the private Chat Completions function representation of a Responses web_search tool
• Web Search Chat Completions — Accepts OpenAI Responses requests with hosted web search while targeting a backend that only implements /v1/chat/completions
• Web Search Scoped Credentials — A scoped-credentials variant of web-search-chat-completions.yaml
• Web Search — Demonstrates the openai_web_search filter configuration