{ "openapi": "3.1.0", "info": { "title": "Palette API", "version": "2026-10-20", "summary": "OpenAI-compatible chat completions with reasoning_effort, served by Palette.", "description": "Base URL https://llm.pltt.dev/v1. Seven routes: GET /v1/health (no auth); GET /v1/models, POST /v1/chat/completions, POST /v1/decision/systemone, GET /v1/decision/models, POST /v1/ocr, GET /v1/ocr/models (Bearer key). Chat is the OpenAI chat-completions shape plus reasoning_effort (low | medium | high | xhigh); reasoning is returned in reasoning_content separate from content. palette-island answers tool calls only (send `tools`). Decision and OCR are Palette services proxied behind the same key. Every field, status code and string here is what the gateway does on 2026-10-20.", "contact": { "name": "Palette", "url": "https://www.pltt.xyz" }, "license": { "name": "Apache 2.0", "identifier": "Apache-2.0" } }, "servers": [ { "url": "https://llm.pltt.dev", "description": "Palette gateway. Routes are under /v1." } ], "security": [ { "bearerAuth": [] } ], "tags": [ { "name": "health", "description": "Upstream status per seat. No key needed." }, { "name": "chat", "description": "Chat completions with reasoning_effort, non-stream or Server-Sent Events." }, { "name": "models", "description": "The five seats, one alias and the two sibling services behind the model field." }, { "name": "decision", "description": "Palette Decision: typed questions with calibrated probabilities." }, { "name": "ocr", "description": "Palette Light OCR: Korean line OCR on one image." } ], "paths": { "/v1/health": { "get": { "tags": [ "health" ], "operationId": "getHealth", "summary": "Health of every seat", "description": "Probes each seat's upstream /v1/models with a 20 s timeout. Always returns 200; a failing seat changes the body, not the status code. ok is true only when every configured seat returned 200.", "security": [], "responses": { "200": { "description": "Per-seat upstream status.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Health" }, "examples": { "allUp": { "summary": "Every seat answers 200", "value": { "ok": true, "seats": { "palette-llm": 200, "palette-llm-max": 200, "palette-vlm": 200, "palette-llm-ko": 200 } } }, "oneDown": { "summary": "One seat timed out", "value": { "ok": false, "seats": { "palette-llm": 200, "palette-llm-max": 200, "palette-vlm": 200, "palette-llm-ko": "ConnectTimeout" } } } } } } } } } }, "/v1/models": { "get": { "tags": [ "models" ], "operationId": "listModels", "summary": "List seats and aliases", "description": "Lists the five seats (palette-llm, palette-llm-max, palette-vlm, palette-llm-ko, palette-island) with tier, configured flag and modalities; chat seats carry the accepted reasoning_effort values, palette-island carries requires: tools. Then the alias palette-llm-fast and two service entries, palette-decision (/v1/decision/systemone) and palette-light-ocr (/v1/ocr), reachable with the same key.", "responses": { "200": { "description": "Model list.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/ModelList" }, "example": { "object": "list", "data": [ { "id": "palette-llm", "object": "model", "owned_by": "palette", "backing": "Palette LLM", "configured": true, "modalities": [ "text" ], "reasoning_effort": [ "low", "medium", "high", "xhigh" ] }, { "id": "palette-llm-max", "object": "model", "owned_by": "palette", "backing": "Palette LLM Max", "configured": true, "modalities": [ "text", "image" ], "reasoning_effort": [ "low", "medium", "high", "xhigh" ] }, { "id": "palette-vlm", "object": "model", "owned_by": "palette", "backing": "Palette Vision", "configured": true, "modalities": [ "text", "image" ], "reasoning_effort": [ "low", "medium", "high", "xhigh" ] }, { "id": "palette-llm-ko", "object": "model", "owned_by": "palette", "backing": "palette-llm", "configured": true, "modalities": [ "text", "image" ], "reasoning_effort": [ "low", "medium", "high", "xhigh" ] }, { "id": "palette-llm-fast", "object": "model", "owned_by": "palette", "alias_of": "palette-vlm" } ] } } } }, "401": { "$ref": "#/components/responses/Unauthorized" } } } }, "/v1/chat/completions": { "post": { "tags": [ "chat" ], "operationId": "createChatCompletion", "summary": "Create a chat completion", "description": "OpenAI chat-completions shape plus reasoning_effort. model omitted resolves to palette-llm; palette-llm-fast resolves to palette-vlm; any other unknown model silently falls back to palette-llm (no 404). reasoning_effort defaults to medium; unknown values silently become medium. When max_tokens is omitted the gateway sets the per-effort default 1024 / 2048 / 8192 / 16384. Fields the gateway does not know (temperature, top_p, stop, n, seed, tools, tool_choice, response_format, stream_options) are forwarded to the backing server untouched. On palette-llm-ko the gateway appends one Korean system message at the end of messages. With stream: true the response is text/event-stream: one line 'data: {json}' per event, a blank line between events, and a final 'data: [DONE]'. On the shared seats stream chunks are forwarded unchanged, so model in a chunk is the backing id and there is no top-level reasoning_effort; do not key on it. xhigh runs two passes (draft, then a self-review with max_tokens 4096); both count against the 900 s request timeout.", "requestBody": { "required": true, "content": { "application/json": { "schema": { "$ref": "#/components/schemas/ChatCompletionRequest" }, "examples": { "low": { "summary": "Non-stream, low effort", "value": { "model": "palette-llm", "reasoning_effort": "low", "messages": [ { "role": "user", "content": "공급사 A(단가 낮음, 납기 2주 지연)와 B(단가 8% 높음, 납기 준수) 중 하나를 고르고 이유를 두 문장으로 답하세요." } ] } }, "streamHigh": { "summary": "Stream, high effort, explicit budget", "value": { "model": "palette-llm", "reasoning_effort": "high", "max_tokens": 4000, "stream": true, "messages": [ { "role": "user", "content": "공급사 A(단가 낮음, 납기 2주 지연)와 B(단가 8% 높음, 납기 준수) 중 하나를 고르고 이유를 두 문장으로 답하세요." } ] } }, "image": { "summary": "Image content part on a vision seat", "value": { "model": "palette-vlm", "reasoning_effort": "low", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "이 세금계산서의 합계 금액을 읽고, 품목 금액의 합과 일치하는지 확인해 주세요." }, { "type": "image_url", "image_url": { "url": "data:image/png;base64,iVBORw0KGgo..." } } ] } ] } } } } } }, "responses": { "200": { "description": "A chat completion. application/json when stream is false or omitted; text/event-stream when stream is true.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/ChatCompletion" }, "example": { "id": "chatcmpl-4f1d0b6c9a2e", "object": "chat.completion", "created": 1790035200, "model": "palette-llm", "reasoning_effort": "low", "choices": [ { "index": 0, "message": { "role": "assistant", "reasoning_content": "A는 단가가 낮지만 납기가 2주 늦어 생산 일정에 영향을 준다. B는 단가가 8% 높지만 납기를 지킨다.", "content": "공급사 B를 선택합니다. 8%의 단가 차이보다 2주 납기 지연으로 인한 생산 중단 비용이 더 크기 때문입니다." }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": 96, "completion_tokens": 88, "total_tokens": 184 } } }, "text/event-stream": { "schema": { "type": "string", "description": "Server-Sent Events. Each event is one line 'data: ' followed by a blank line; the stream ends with 'data: [DONE]'. Reasoning arrives in choices[0].delta.reasoning_content, the answer in choices[0].delta.content. Some upstreams send a final chunk with empty choices carrying usage. Chunks the gateway synthesizes may omit created or usage; always read choices[0].delta.", "x-event-schema": { "$ref": "#/components/schemas/ChatCompletionChunk" } }, "example": "data: {\"id\":\"chatcmpl-9b7e2c1a\",\"object\":\"chat.completion.chunk\",\"created\":1790035201,\"model\":\"palette-llm-ko\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"reasoning_content\":\"공급사 A는 단가가 낮지만 \"},\"finish_reason\":null}]}\n\ndata: {\"id\":\"chatcmpl-9b7e2c1a\",\"object\":\"chat.completion.chunk\",\"created\":1790035201,\"model\":\"palette-llm-ko\",\"choices\":[{\"index\":0,\"delta\":{\"reasoning_content\":\"납기가 2주 늦다. B는 8% 비싸지만 납기를 지킨다.\"},\"finish_reason\":null}]}\n\ndata: {\"id\":\"chatcmpl-9b7e2c1a\",\"object\":\"chat.completion.chunk\",\"created\":1790035201,\"model\":\"palette-llm-ko\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"공급사 B를 선택합니다.\"},\"finish_reason\":null}]}\n\ndata: {\"id\":\"chatcmpl-9b7e2c1a\",\"object\":\"chat.completion.chunk\",\"created\":1790035201,\"model\":\"palette-llm-ko\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\" 납기 지연 비용이 단가 차이보다 큽니다.\"},\"finish_reason\":null}]}\n\ndata: {\"id\":\"chatcmpl-9b7e2c1a\",\"object\":\"chat.completion.chunk\",\"created\":1790035201,\"model\":\"palette-llm-ko\",\"choices\":[{\"index\":0,\"delta\":{},\"finish_reason\":\"stop\"}]}\n\ndata: [DONE]\n" } } }, "401": { "$ref": "#/components/responses/Unauthorized" }, "503": { "description": "The seat has no upstream URL configured, or the Palette cloud proxy token is missing (shared seats).", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" }, "examples": { "noUpstream": { "value": { "detail": "seat has no upstream configured yet" } }, "noProxyToken": { "value": { "detail": "Palette cloud proxy token not configured" } } } } } }, "default": { "description": "Upstream 4xx or 5xx passed through with the same status code; detail is the first 500 characters of the upstream body. Separately, Cloudflare returns 403 'error code: 1010' (HTML or text, not JSON) to Python's stock urllib User-Agent only; the OpenAI SDK, requests, httpx, curl and fetch all pass.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" } }, "text/html": { "schema": { "type": "string" } } } } } } }, "/v1/decision/systemone": { "post": { "tags": [ "decision" ], "operationId": "decide", "summary": "Typed decision over one state", "description": "Forwards the body untouched to Palette Decision and returns its answer. Up to 32 questions, scored independently. Models: palette-decision-27b (K-DECIDE 800 items 0.944, training-harness eval 2026-10-15) and the default palette-decision-latest (served 7.3B, 0.873 on the same items, faster). Measured through the gateway: two-question 27B request 1.1 s.", "requestBody": { "required": true, "content": { "application/json": { "schema": { "$ref": "#/components/schemas/DecisionRequest" } } } }, "responses": { "200": { "description": "Answers per question.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/DecisionResponse" } } } }, "401": { "description": "Error.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" } } } }, "503": { "description": "Error.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" } } } } } } }, "/v1/decision/models": { "get": { "tags": [ "decision" ], "operationId": "listDecisionModels", "summary": "Decision model ids", "responses": { "200": { "description": "Model list.", "content": { "application/json": { "example": { "data": [ { "id": "palette-decision-latest", "object": "model", "owned_by": "palette" }, { "id": "palette-decision-27b", "object": "model", "owned_by": "palette" } ] } } } }, "401": { "description": "Error.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" } } } } } } }, "/v1/ocr": { "post": { "tags": [ "ocr" ], "operationId": "ocr", "summary": "Korean line OCR on one image", "description": "Send the image as multipart `file`, as JSON `image_b64` (data-URL prefix accepted), or as raw bytes with an image Content-Type. 20 MB limit; long side downscaled to 4000 px; EXIF rotation applied. Lines come back in reading order. Measured: 48 lines on an A4 test page in 1.6 s server-side; recogniser CER 0.08% on 600 held-out synthetic Korean lines; real scans and photos not yet measured.", "parameters": [ { "name": "lang", "in": "query", "schema": { "type": "string", "default": "ko" } }, { "name": "min_conf", "in": "query", "schema": { "type": "number", "default": 0 } } ], "requestBody": { "required": true, "content": { "multipart/form-data": { "schema": { "type": "object", "properties": { "file": { "type": "string", "format": "binary" } } } }, "application/json": { "schema": { "$ref": "#/components/schemas/OcrJsonRequest" } }, "image/*": { "schema": { "type": "string", "format": "binary" } } } }, "responses": { "200": { "description": "Lines and joined text.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/OcrResponse" } } } }, "400": { "description": "Error.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" } } } }, "401": { "description": "Error.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" } } } }, "413": { "description": "Error.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" } } } }, "503": { "description": "Error.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" } } } } } } }, "/v1/ocr/models": { "get": { "tags": [ "ocr" ], "operationId": "listOcrModels", "summary": "OCR model, version and languages", "responses": { "200": { "description": "Model list.", "content": { "application/json": { "example": { "object": "list", "data": [ { "id": "palette-light-ocr", "object": "model", "owned_by": "palette", "version": "palette-light-ocr-tiny-v0", "languages": [ "ko", "en", "digits", "symbols" ] } ] } } } }, "401": { "description": "Error.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" } } } } } } } }, "components": { "securitySchemes": { "bearerAuth": { "type": "http", "scheme": "bearer", "bearerFormat": "pk_ followed by 48 URL-safe characters", "description": "Authorization: Bearer on every route except GET /v1/health. Keys are minted by the operator and stored server-side as a SHA-256 hash. A revoked key can keep working for up to 60 seconds (per-container cache TTL). No per-key quotas are enforced today." } }, "responses": { "Unauthorized": { "description": "Missing Authorization header, or an unknown or revoked key.", "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Error" }, "examples": { "missingHeader": { "summary": "No Authorization: Bearer header", "value": { "detail": "Authorization: Bearer 헤더가 필요합니다." } }, "badKey": { "summary": "Unknown or revoked key", "value": { "detail": "유효하지 않거나 폐기된 키입니다." } } } } } } }, "schemas": { "Seat": { "type": "string", "description": "Seat id or alias. palette-llm: Palette LLM (Palette LLM), text only, 1M context, default. palette-llm-max: Palette LLM Max (Palette LLM Max), vision, 1M. palette-vlm: Palette Vision (Palette Vision), vision, 1M, alias palette-llm-fast. palette-llm-ko: Palette LLM Korean on Palette's own GPU, vision (up to 6 images per request), 256K. Any other string silently falls back to palette-llm. palette-island answers tool calls only: send an OpenAI `tools` array or the gateway returns 400.", "enum": [ "palette-llm", "palette-llm-max", "palette-vlm", "palette-llm-ko", "palette-llm-fast", "palette-island" ], "default": "palette-llm" }, "ReasoningEffort": { "type": "string", "description": "Case-insensitive; '-' and '_' are stripped. minimal becomes low; extra-high, extrahigh and max become xhigh; anything unrecognized silently becomes medium. Shared seats map low and medium to native low, high to native high, xhigh to native max plus a self-review pass. palette-llm-ko maps them to a Korean reasoning protocol. Default max_tokens when omitted: low 1024, medium 2048, high 8192, xhigh 16384 (draft) with the review pass fixed at 4096.", "enum": [ "low", "medium", "high", "xhigh" ], "default": "medium" }, "TextPart": { "type": "object", "required": [ "type", "text" ], "properties": { "type": { "const": "text" }, "text": { "type": "string" } } }, "ImageUrlPart": { "type": "object", "description": "Accepted on palette-vlm (alias palette-llm-fast), palette-llm-max and palette-llm-ko. palette-llm is text only. palette-llm-ko accepts up to 6 images per request.", "required": [ "type", "image_url" ], "properties": { "type": { "const": "image_url" }, "image_url": { "type": "object", "required": [ "url" ], "properties": { "url": { "type": "string", "description": "data:image/png;base64,... or an https:// URL." } } } } }, "ContentPart": { "oneOf": [ { "$ref": "#/components/schemas/TextPart" }, { "$ref": "#/components/schemas/ImageUrlPart" } ], "discriminator": { "propertyName": "type", "mapping": { "text": "#/components/schemas/TextPart", "image_url": "#/components/schemas/ImageUrlPart" } } }, "Message": { "type": "object", "required": [ "role", "content" ], "properties": { "role": { "type": "string", "enum": [ "system", "user", "assistant" ] }, "content": { "description": "A string, or an array of parts (text and, on vision seats, image_url).", "oneOf": [ { "type": "string" }, { "type": "array", "items": { "$ref": "#/components/schemas/ContentPart" } } ] } } }, "ChatCompletionRequest": { "type": "object", "required": [ "messages" ], "additionalProperties": true, "description": "OpenAI chat-completions body plus reasoning_effort. Unknown fields are forwarded to the backing server untouched; the gateway neither validates nor guarantees them.", "properties": { "model": { "$ref": "#/components/schemas/Seat" }, "messages": { "type": "array", "minItems": 1, "items": { "$ref": "#/components/schemas/Message" }, "description": "OpenAI messages. On palette-llm-ko the gateway appends one Korean system message (the effort instruction) at the end." }, "reasoning_effort": { "$ref": "#/components/schemas/ReasoningEffort" }, "reasoning": { "type": "object", "description": "Alternative spelling. Wins over reasoning_effort when both are present. Consumed by the gateway, never forwarded verbatim.", "properties": { "effort": { "$ref": "#/components/schemas/ReasoningEffort" } } }, "max_tokens": { "type": "integer", "minimum": 1, "description": "Forwarded. When omitted the gateway sets the per-effort default: 1024 / 2048 / 8192 / 16384. Set at least 4000 for high and xhigh; a small budget can be consumed by thinking and leave content empty." }, "stream": { "type": "boolean", "default": false, "description": "true returns Server-Sent Events (text/event-stream)." }, "temperature": { "type": "number", "description": "Forwarded untouched." }, "top_p": { "type": "number", "description": "Forwarded untouched." }, "stop": { "description": "Forwarded untouched. Support depends on the backing model's server.", "oneOf": [ { "type": "string" }, { "type": "array", "items": { "type": "string" } } ] } } }, "AssistantMessage": { "type": "object", "required": [ "role", "content" ], "properties": { "role": { "const": "assistant" }, "content": { "type": [ "string", "null" ], "description": "The answer. Can be empty when a small max_tokens was consumed by thinking; render reasoning_content in that case." }, "reasoning_content": { "type": "string", "description": "Thinking, separate from the answer. Added by the gateway. Empty string when the model produced none. At xhigh it also carries '[초안 답] ' and '[검토] '." } } }, "Choice": { "type": "object", "required": [ "index", "message" ], "properties": { "index": { "type": "integer" }, "message": { "$ref": "#/components/schemas/AssistantMessage" }, "finish_reason": { "type": [ "string", "null" ] } } }, "Usage": { "type": "object", "properties": { "prompt_tokens": { "type": "integer" }, "completion_tokens": { "type": "integer", "description": "Includes thinking tokens." }, "total_tokens": { "type": "integer" } } }, "ChatCompletion": { "type": "object", "required": [ "id", "object", "model", "choices" ], "description": "Non-stream response. model (seat id), reasoning_effort (normalized) and choices[0].message.reasoning_content are set by the gateway; every other field comes from the upstream.", "properties": { "id": { "type": "string" }, "object": { "const": "chat.completion" }, "created": { "type": "integer", "description": "Unix seconds." }, "model": { "$ref": "#/components/schemas/Seat" }, "reasoning_effort": { "$ref": "#/components/schemas/ReasoningEffort" }, "choices": { "type": "array", "items": { "$ref": "#/components/schemas/Choice" } }, "usage": { "$ref": "#/components/schemas/Usage" } } }, "Delta": { "type": "object", "properties": { "role": { "const": "assistant", "description": "First chunk only." }, "content": { "type": "string", "description": "A piece of the answer." }, "reasoning_content": { "type": "string", "description": "A piece of the thinking. Arrives before content." } } }, "ChunkChoice": { "type": "object", "required": [ "index", "delta" ], "properties": { "index": { "type": "integer" }, "delta": { "$ref": "#/components/schemas/Delta" }, "finish_reason": { "type": [ "string", "null" ] } } }, "ChatCompletionChunk": { "type": "object", "required": [ "id", "object", "choices" ], "description": "One 'data:' event of a stream. On the shared seats chunks are the upstream's bytes: model is the backing id (for example Palette LLM) and there is no top-level reasoning_effort. On palette-llm-ko at high and xhigh the gateway re-emits chunks with model set to the seat id and routes text around the [답변] marker. Synthesized chunks may omit created or usage. Some upstreams send a final chunk with empty choices carrying usage; guard for it.", "properties": { "id": { "type": "string" }, "object": { "const": "chat.completion.chunk" }, "created": { "type": "integer" }, "model": { "type": "string", "description": "Backing id on the shared seats, seat id on palette-llm-ko high and xhigh. Do not key on it." }, "choices": { "type": "array", "items": { "$ref": "#/components/schemas/ChunkChoice" } }, "usage": { "$ref": "#/components/schemas/Usage" } } }, "Model": { "type": "object", "required": [ "id", "object", "owned_by" ], "description": "A seat entry carries backing, configured, modalities and reasoning_effort. An alias entry carries alias_of and no other capability fields.", "properties": { "id": { "$ref": "#/components/schemas/Seat" }, "object": { "const": "model" }, "owned_by": { "const": "palette" }, "backing": { "type": "string", "description": "Model id the gateway sends upstream. For palette-llm-ko this is the Palette inference server served-model name (palette-llm by default), not the Hugging Face id." }, "configured": { "type": "boolean", "description": "An upstream URL is set." }, "modalities": { "type": "array", "items": { "type": "string", "enum": [ "text", "image" ] } }, "reasoning_effort": { "type": "array", "items": { "$ref": "#/components/schemas/ReasoningEffort" } }, "alias_of": { "$ref": "#/components/schemas/Seat" } } }, "ModelList": { "type": "object", "required": [ "object", "data" ], "properties": { "object": { "const": "list" }, "data": { "type": "array", "items": { "$ref": "#/components/schemas/Model" } } } }, "Health": { "type": "object", "required": [ "ok", "seats" ], "properties": { "ok": { "type": "boolean", "description": "true only when every configured seat returned 200." }, "seats": { "type": "object", "description": "One entry per seat: the upstream HTTP status as an integer, the string 'unconfigured' when no upstream URL is set, or an exception class name such as 'ConnectTimeout'.", "additionalProperties": { "oneOf": [ { "type": "integer" }, { "type": "string" } ] } } } }, "Error": { "type": "object", "required": [ "detail" ], "properties": { "detail": { "type": "string", "description": "FastAPI-shaped message. Korean for the two 401 cases." } } }, "DecisionQuestion": { "type": "object", "required": [ "type", "instructions" ], "properties": { "type": { "type": "string", "enum": [ "choice", "noul", "score" ] }, "instructions": { "type": "string", "description": "The question, usually Korean." }, "criteria": { "description": "choice: object option -> description (or null). score: ordered list of level descriptions, low to high. noul: optional {\"true\": ..., \"false\": ...}.", "oneOf": [ { "type": "object", "additionalProperties": { "type": [ "string", "null" ] } }, { "type": "array", "items": { "type": "string" }, "minItems": 2, "maxItems": 10 } ] } } }, "DecisionRequest": { "type": "object", "required": [ "state", "questions" ], "properties": { "model": { "type": "string", "default": "palette-decision-latest", "enum": [ "palette-decision-latest", "palette-decision-27b" ] }, "state": { "description": "Text or JSON object describing the situation.", "oneOf": [ { "type": "string" }, { "type": "object" } ] }, "questions": { "type": "object", "minProperties": 1, "maxProperties": 32, "additionalProperties": { "$ref": "#/components/schemas/DecisionQuestion" } } }, "example": { "model": "palette-decision-27b", "state": "배송이 3일째 물류센터에 멈춰 있어요. 오늘 안에 출발 안 되면 취소하고 환불받을게요.", "questions": { "team": { "type": "choice", "instructions": "이 문의를 처리할 팀은?", "criteria": { "배송": "배송 지연·추적", "환불": "취소·환불 처리", "기타": "그 외" } }, "cancel": { "type": "noul", "instructions": "고객이 취소 의사를 밝혔는가?" }, "urgency": { "type": "score", "instructions": "긴급도를 평가하라.", "criteria": [ "기한 없음", "며칠 내", "명확한 기한 있음", "오늘 안에 처리 필요" ] } } } }, "DecisionAnswer": { "type": "object", "required": [ "type", "probabilities", "confidence" ], "properties": { "type": { "type": "string", "enum": [ "choice", "noul", "score" ] }, "choice": { "type": "string" }, "noul": { "type": "number" }, "score": { "type": "number" }, "probabilities": { "type": "object", "additionalProperties": { "type": "number" } }, "confidence": { "type": "number" }, "legend": { "type": "object", "additionalProperties": { "type": "string" } } } }, "DecisionResponse": { "type": "object", "required": [ "model", "answers" ], "properties": { "model": { "type": "string" }, "answers": { "type": "object", "additionalProperties": { "$ref": "#/components/schemas/DecisionAnswer" } }, "usage": { "type": "object", "properties": { "input_tokens": { "type": "integer" }, "shared_prefix_tokens": { "type": "integer" }, "output_tokens": { "type": "integer" } } } } }, "OcrLine": { "type": "object", "required": [ "text", "confidence", "box" ], "properties": { "text": { "type": "string" }, "confidence": { "type": "number" }, "box": { "type": "array", "minItems": 4, "maxItems": 4, "items": { "type": "array", "minItems": 2, "maxItems": 2, "items": { "type": "integer" } } } } }, "OcrResponse": { "type": "object", "required": [ "model", "text", "lines" ], "properties": { "model": { "type": "string", "example": "palette-light-ocr" }, "lang": { "type": "string" }, "text": { "type": "string" }, "lines": { "type": "array", "items": { "$ref": "#/components/schemas/OcrLine" } }, "width": { "type": "integer" }, "height": { "type": "integer" }, "ms": { "type": "object", "properties": { "detect": { "type": "number" }, "recognise": { "type": "number" }, "total": { "type": "number" } } } } }, "OcrJsonRequest": { "type": "object", "properties": { "image_b64": { "type": "string", "description": "Base64 image, plain or data URL (alias: image)." } } } } } }