mirror of
https://github.com/diegosouzapw/OmniRoute.git
synced 2026-08-06 07:12:12 +03:00
Adds intelligent cache control preservation for Claude Code clients:
- New cacheControlPolicy.ts module with detection logic:
- isClaudeCodeClient(): Detects Claude Code via User-Agent
- providerSupportsCaching(): Checks provider (claude, anthropic, zai, qwen)
- isDeterministicStrategy(): Identifies priority/cost-optimized strategies
- shouldPreserveCacheControl(): Main policy decision
- Cache control is preserved when:
1. Client is Claude Code (detected via User-Agent)
2. Provider supports prompt caching
3. Request routing is deterministic:
- Single model requests (always)
- Combo with priority or cost-optimized strategy only
- Updated translator to accept preserveCacheControl option
- Updated chatCore and chat handler to propagate combo strategy
- Added comprehensive unit tests (24 tests)
Non-deterministic combo strategies (weighted, round-robin, random, etc.)
continue to use OmniRoute's managed caching strategy.
Refs: #cache-control-preservation
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
338 lines
11 KiB
TypeScript
338 lines
11 KiB
TypeScript
import { FORMATS } from "./formats.ts";
|
|
import { ensureToolCallIds, fixMissingToolResponses } from "./helpers/toolCallHelper.ts";
|
|
import { prepareClaudeRequest } from "./helpers/claudeHelper.ts";
|
|
import { filterToOpenAIFormat } from "./helpers/openaiHelper.ts";
|
|
import {
|
|
coerceToolSchemas,
|
|
injectEmptyReasoningContentForToolCalls,
|
|
sanitizeToolDescriptions,
|
|
} from "./helpers/schemaCoercion.ts";
|
|
import { getRequestTranslator, getResponseTranslator } from "./registry.ts";
|
|
import { bootstrapTranslatorRegistry } from "./bootstrap.ts";
|
|
import { normalizeThinkingConfig } from "../services/provider.ts";
|
|
import { applyThinkingBudget } from "../services/thinkingBudget.ts";
|
|
import { normalizeRoles } from "../services/roleNormalizer.ts";
|
|
|
|
bootstrapTranslatorRegistry();
|
|
export { register } from "./registry.ts";
|
|
|
|
function normalizeResponsesInputItem(item) {
|
|
if (typeof item === "string") {
|
|
return {
|
|
type: "message",
|
|
role: "user",
|
|
content: [{ type: "input_text", text: item }],
|
|
};
|
|
}
|
|
|
|
if (!item || typeof item !== "object") return item;
|
|
|
|
if (item.type || item.role) {
|
|
return item.type ? item : { type: "message", ...item };
|
|
}
|
|
|
|
if (typeof item.text === "string") {
|
|
return {
|
|
type: "message",
|
|
role: "user",
|
|
content: [{ type: "input_text", text: item.text }],
|
|
};
|
|
}
|
|
|
|
return item;
|
|
}
|
|
|
|
function normalizeOpenAIResponsesRequest(body) {
|
|
if (!body || typeof body !== "object") return body;
|
|
|
|
const normalized = { ...body };
|
|
|
|
if (typeof normalized.input === "string") {
|
|
normalized.input = [
|
|
{
|
|
type: "message",
|
|
role: "user",
|
|
content: [{ type: "input_text", text: normalized.input }],
|
|
},
|
|
];
|
|
return normalized;
|
|
}
|
|
|
|
if (Array.isArray(normalized.input)) {
|
|
normalized.input = normalized.input.map(normalizeResponsesInputItem);
|
|
return normalized;
|
|
}
|
|
|
|
if (normalized.input && typeof normalized.input === "object") {
|
|
normalized.input = [normalizeResponsesInputItem(normalized.input)];
|
|
return normalized;
|
|
}
|
|
|
|
return normalized;
|
|
}
|
|
|
|
/** @param options.normalizeToolCallId - When true, use 9-char tool call ids (e.g. Mistral); when false, leave ids as-is */
|
|
/** @param options.preserveDeveloperRole - undefined/true: keep developer for OpenAI format (default); false: map to system */
|
|
/** @param options.preserveCacheControl - When true, preserve client-side cache_control markers (for Claude Code, etc.) */
|
|
// Translate request: source -> openai -> target
|
|
export function translateRequest(
|
|
sourceFormat,
|
|
targetFormat,
|
|
model,
|
|
body,
|
|
stream = true,
|
|
credentials = null,
|
|
provider = null,
|
|
reqLogger = null,
|
|
options?: { normalizeToolCallId?: boolean; preserveDeveloperRole?: boolean; preserveCacheControl?: boolean }
|
|
) {
|
|
let result = body;
|
|
const use9CharId = options?.normalizeToolCallId === true;
|
|
const preserveDeveloperRole = options?.preserveDeveloperRole;
|
|
|
|
// Phase 2: Apply thinking budget control before normalization
|
|
result = applyThinkingBudget(result);
|
|
|
|
// Normalize thinking config: remove if lastMessage is not user
|
|
normalizeThinkingConfig(result);
|
|
|
|
// Ensure tool_calls have id; optionally normalize to 9-char for providers like Mistral
|
|
ensureToolCallIds(result, { use9CharId });
|
|
|
|
// Fix missing tool responses (insert empty tool_result if needed)
|
|
fixMissingToolResponses(result);
|
|
|
|
// Normalize roles: developer→system unless preserved, system→user for incompatible models.
|
|
// This handles (1) sourceFormat openai with messages containing developer → non-openai target
|
|
// or preserveDeveloperRole=false, and (2) all other paths where result.messages already exists.
|
|
if (result.messages && Array.isArray(result.messages)) {
|
|
result.messages = normalizeRoles(
|
|
result.messages,
|
|
provider || "",
|
|
model || "",
|
|
targetFormat,
|
|
preserveDeveloperRole
|
|
);
|
|
}
|
|
|
|
// If same format, skip translation steps
|
|
if (sourceFormat !== targetFormat) {
|
|
// Check for direct translation path first (e.g., Claude → Gemini)
|
|
const directTranslator = getRequestTranslator(sourceFormat, targetFormat);
|
|
if (directTranslator && sourceFormat !== FORMATS.OPENAI && targetFormat !== FORMATS.OPENAI) {
|
|
result = directTranslator(model, result, stream, credentials);
|
|
} else {
|
|
// Fallback: hub-and-spoke via OpenAI
|
|
// Step 1: source -> openai (if source is not openai)
|
|
if (sourceFormat !== FORMATS.OPENAI) {
|
|
const toOpenAI = getRequestTranslator(sourceFormat, FORMATS.OPENAI);
|
|
if (toOpenAI) {
|
|
result = toOpenAI(model, result, stream, credentials);
|
|
// Log OpenAI intermediate format
|
|
reqLogger?.logOpenAIRequest?.(result);
|
|
}
|
|
}
|
|
|
|
// Step 2: openai -> target (if target is not openai)
|
|
if (targetFormat !== FORMATS.OPENAI) {
|
|
const fromOpenAI = getRequestTranslator(FORMATS.OPENAI, targetFormat);
|
|
if (fromOpenAI) {
|
|
result = fromOpenAI(model, result, stream, credentials);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// Always normalize to clean OpenAI format when target is OpenAI
|
|
// This handles hybrid requests (e.g., OpenAI messages + Claude tools)
|
|
if (targetFormat === FORMATS.OPENAI) {
|
|
result = filterToOpenAIFormat(result);
|
|
}
|
|
|
|
// Final step: prepare request for Claude format endpoints
|
|
// Preserve cache_control when:
|
|
// 1. Claude passthrough mode (Claude → Claude), OR
|
|
// 2. Explicitly requested via options (for caching-aware clients like Claude Code)
|
|
if (targetFormat === FORMATS.CLAUDE) {
|
|
const isClaudePassthrough = sourceFormat === FORMATS.CLAUDE;
|
|
const preserveCache = isClaudePassthrough || options?.preserveCacheControl === true;
|
|
result = prepareClaudeRequest(result, provider, preserveCache);
|
|
}
|
|
|
|
// Normalize openai-responses input shape for providers that require list input.
|
|
if (targetFormat === FORMATS.OPENAI_RESPONSES) {
|
|
result = normalizeOpenAIResponsesRequest(result);
|
|
}
|
|
|
|
// Second role normalization: only for OPENAI_RESPONSES. Here messages are built from input
|
|
// after the translation step, so the first normalizeRoles (above) did not see them. For
|
|
// sourceFormat openai with messages already on the body, the first block handles developer
|
|
// → system (non-openai target or preserveDeveloperRole=false); no second pass needed.
|
|
if (
|
|
sourceFormat === FORMATS.OPENAI_RESPONSES &&
|
|
result.messages &&
|
|
Array.isArray(result.messages)
|
|
) {
|
|
result.messages = normalizeRoles(
|
|
result.messages,
|
|
provider || "",
|
|
model || "",
|
|
targetFormat,
|
|
preserveDeveloperRole
|
|
);
|
|
}
|
|
|
|
if (result.tools !== undefined) {
|
|
result.tools = coerceToolSchemas(result.tools);
|
|
result.tools = sanitizeToolDescriptions(result.tools);
|
|
}
|
|
|
|
if (targetFormat === FORMATS.OPENAI && result.messages && Array.isArray(result.messages)) {
|
|
result.messages = injectEmptyReasoningContentForToolCalls(result.messages, provider);
|
|
}
|
|
|
|
// Ensure unique tool_call ids on final payload (translators may have introduced duplicates)
|
|
ensureToolCallIds(result, { use9CharId });
|
|
fixMissingToolResponses(result);
|
|
|
|
if (result.tools) {
|
|
result.tools = coerceToolSchemas(result.tools);
|
|
result.tools = sanitizeToolDescriptions(result.tools);
|
|
}
|
|
|
|
// Inject reasoning_content = "" for DeepSeek/Reasoning models assistant messages with tool_calls
|
|
// if omitted by the client, to avoid upstream 400 errors (e.g. "Messages with role 'assistant' that contain tool_calls must also include reasoning_content")
|
|
const isReasoner =
|
|
provider === "deepseek" || (typeof model === "string" && /r1|reason/i.test(model));
|
|
if (isReasoner && result.messages && Array.isArray(result.messages)) {
|
|
for (const msg of result.messages) {
|
|
if (
|
|
msg.role === "assistant" &&
|
|
Array.isArray(msg.tool_calls) &&
|
|
msg.tool_calls.length > 0 &&
|
|
msg.reasoning_content === undefined
|
|
) {
|
|
msg.reasoning_content = "";
|
|
}
|
|
}
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
// Translate response chunk: target -> openai -> source
|
|
export function translateResponse(targetFormat, sourceFormat, chunk, state) {
|
|
// If same format, return as-is
|
|
if (sourceFormat === targetFormat) {
|
|
return [chunk];
|
|
}
|
|
|
|
let results = [chunk];
|
|
let openaiResults = null; // Store OpenAI intermediate results
|
|
|
|
// Check for direct translation path first (e.g., Gemini → Claude)
|
|
const directTranslator = getResponseTranslator(targetFormat, sourceFormat);
|
|
if (directTranslator && targetFormat !== FORMATS.OPENAI && sourceFormat !== FORMATS.OPENAI) {
|
|
const converted = directTranslator(chunk, state);
|
|
if (converted) {
|
|
results = Array.isArray(converted) ? converted : [converted];
|
|
} else {
|
|
results = [];
|
|
}
|
|
return results;
|
|
}
|
|
|
|
// Fallback: hub-and-spoke via OpenAI
|
|
// Step 1: target -> openai (if target is not openai)
|
|
if (targetFormat !== FORMATS.OPENAI) {
|
|
const toOpenAI = getResponseTranslator(targetFormat, FORMATS.OPENAI);
|
|
if (toOpenAI) {
|
|
results = [];
|
|
const converted = toOpenAI(chunk, state);
|
|
if (converted) {
|
|
results = Array.isArray(converted) ? converted : [converted];
|
|
openaiResults = results; // Store OpenAI intermediate
|
|
}
|
|
}
|
|
}
|
|
|
|
// Step 2: openai -> source (if source is not openai)
|
|
if (sourceFormat !== FORMATS.OPENAI) {
|
|
const fromOpenAI = getResponseTranslator(FORMATS.OPENAI, sourceFormat);
|
|
if (fromOpenAI) {
|
|
const finalResults = [];
|
|
for (const r of results) {
|
|
const converted = fromOpenAI(r, state);
|
|
if (converted) {
|
|
finalResults.push(...(Array.isArray(converted) ? converted : [converted]));
|
|
}
|
|
}
|
|
results = finalResults;
|
|
}
|
|
}
|
|
|
|
// Attach OpenAI intermediate results for logging
|
|
if (openaiResults && sourceFormat !== FORMATS.OPENAI && targetFormat !== FORMATS.OPENAI) {
|
|
(results as { _openaiIntermediate?: unknown })._openaiIntermediate = openaiResults;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// Check if translation needed
|
|
export function needsTranslation(sourceFormat, targetFormat) {
|
|
return sourceFormat !== targetFormat;
|
|
}
|
|
|
|
// Initialize state for streaming response based on format
|
|
export function initState(sourceFormat) {
|
|
// Base state for all formats
|
|
const base = {
|
|
messageId: null,
|
|
model: null,
|
|
textBlockStarted: false,
|
|
thinkingBlockStarted: false,
|
|
inThinkingBlock: false,
|
|
currentBlockIndex: null,
|
|
toolCalls: new Map(),
|
|
finishReason: null,
|
|
finishReasonSent: false,
|
|
usage: null,
|
|
contentBlockIndex: -1,
|
|
};
|
|
|
|
// Add openai-responses specific fields
|
|
if (sourceFormat === FORMATS.OPENAI_RESPONSES) {
|
|
return {
|
|
...base,
|
|
seq: 0,
|
|
responseId: `resp_${Date.now()}`,
|
|
created: Math.floor(Date.now() / 1000),
|
|
started: false,
|
|
msgTextBuf: {},
|
|
msgItemAdded: {},
|
|
msgContentAdded: {},
|
|
msgItemDone: {},
|
|
reasoningId: "",
|
|
reasoningIndex: -1,
|
|
reasoningBuf: "",
|
|
reasoningPartAdded: false,
|
|
reasoningDone: false,
|
|
inThinking: false,
|
|
funcArgsBuf: {},
|
|
funcNames: {},
|
|
funcCallIds: {},
|
|
funcArgsDone: {},
|
|
funcItemDone: {},
|
|
completedSent: false,
|
|
};
|
|
}
|
|
|
|
return base;
|
|
}
|
|
|
|
// Initialize all translators (no-op, kept for backward compatibility)
|
|
export function initTranslators() {
|
|
bootstrapTranslatorRegistry();
|
|
}
|