Skip to content

Commit a3853fb

Browse files
committed
Fix OpenCode MissingSessionID 400: add x-opencode-session header
The OpenCode gateway (opencode.ai/zen/go) tightened routing enforcement on 2026-09-06 and now returns HTTP 400 MissingSessionID when the x-opencode-session header is missing. See: https://opencode.ai/docs/go/#where-can-i-use-it Changes: - llm-client.ts: add x-opencode-session header to both callLLM and llmChatCompletion. Accept optional sessionId in both, default to crypto.randomUUID() when caller doesn't supply one. - translation-pipeline.ts: generate one sessionId per pipeline run (runFastTranslation, runTranslationPipeline) and thread it through translate/validate/refine so the gateway can apply prompt-cache affinity across all calls in the same conversation. Refs: anomalyco/opencode#47763 (live repro today).
1 parent ff7cc7e commit a3853fb

2 files changed

Lines changed: 36 additions & 15 deletions

File tree

‎src/lib/llm-client.ts‎

Lines changed: 17 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -14,8 +14,18 @@
1414
* producing the final answer in `content`. The parser also reads
1515
* `reasoning_content` as a safety fallback if the gateway ever returns content
1616
* there.
17+
*
18+
* Session routing:
19+
* Per https://opencode.ai/docs/go/#where-can-i-use-it, the gateway requires an
20+
* `x-opencode-session` header for request routing. Enforcement tightened on
21+
* 2026-09-06; requests without it now return HTTP 400 with
22+
* `MissingSessionID`. We generate a stable UUID per pipeline run and reuse it
23+
* across all calls in the same pipeline so the gateway can apply prompt-cache
24+
* affinity. A fresh UUID is generated if the caller doesn't supply one.
1725
*/
1826

27+
import { randomUUID } from 'node:crypto';
28+
1929
const LLM_BASE_URL = 'https://opencode.ai/zen/go';
2030
const DEFAULT_MODEL = 'glm-5.1';
2131

@@ -30,6 +40,7 @@ export interface LLMChatOptions {
3040
temperature?: number;
3141
maxTokens?: number;
3242
apiKey: string; // Required — always pass explicitly
43+
sessionId?: string; // Stable per-conversation ID for OpenCode routing/cache affinity
3344
}
3445

3546
export interface LLMChatResponse {
@@ -67,6 +78,7 @@ function buildRequestBody(
6778

6879
export async function llmChatCompletion(options: LLMChatOptions): Promise<LLMChatResponse> {
6980
const model = options.model || DEFAULT_MODEL;
81+
const sessionId = options.sessionId || randomUUID();
7082
const controller = new AbortController();
7183
const timeout = setTimeout(() => controller.abort(), 50_000); // 50s — leaves 10s buffer for Vercel's 60s maxDuration
7284

@@ -76,6 +88,7 @@ export async function llmChatCompletion(options: LLMChatOptions): Promise<LLMCha
7688
headers: {
7789
'Content-Type': 'application/json',
7890
'Authorization': `Bearer ${options.apiKey}`,
91+
'x-opencode-session': sessionId,
7992
},
8093
body: JSON.stringify(buildRequestBody(
8194
model,
@@ -129,9 +142,11 @@ export async function callLLM(
129142
apiKey: string,
130143
model?: string,
131144
maxTokens: number = 4096,
132-
temperature: number = 0.3
145+
temperature: number = 0.3,
146+
sessionId?: string
133147
): Promise<string> {
134148
const modelName = model || DEFAULT_MODEL;
149+
const resolvedSessionId = sessionId || randomUUID();
135150
const controller = new AbortController();
136151
const timeout = setTimeout(() => controller.abort(), 50_000); // 50s — leaves 10s buffer for Vercel's 60s maxDuration
137152

@@ -141,6 +156,7 @@ export async function callLLM(
141156
headers: {
142157
'Content-Type': 'application/json',
143158
'Authorization': `Bearer ${apiKey}`,
159+
'x-opencode-session': resolvedSessionId,
144160
},
145161
body: JSON.stringify(buildRequestBody(
146162
modelName,

‎src/lib/translation-pipeline.ts‎

Lines changed: 19 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -21,6 +21,7 @@
2121
*/
2222

2323
import { callLLM, DEFAULT_MODEL } from './llm-client';
24+
import { randomUUID } from 'node:crypto';
2425

2526
// ─── Types ───────────────────────────────────────────────────────────────────
2627

@@ -120,7 +121,7 @@ Source: ${input.text.substring(0, 200)}... → ${input.targetLanguage}`;
120121

121122
// ─── Activity 1: translateText ───────────────────────────────────────────────
122123

123-
async function translateText(input: TranslationRequest, isRetry: boolean = false): Promise<{ translatedText: string; model: string }> {
124+
async function translateText(input: TranslationRequest, isRetry: boolean = false, sessionId?: string): Promise<{ translatedText: string; model: string }> {
124125
const srcLabel = input.sourceLanguage === 'auto' ? 'the detected source language' : input.sourceLanguage;
125126
const targetLabel = input.targetLanguage;
126127

@@ -199,7 +200,7 @@ Rules:
199200

200201
// Use higher temperature for same-language transformation (more creative/structured output)
201202
const temperature = isSameLanguage ? 0.5 : 0.3;
202-
const result = await callLLM(systemPrompt, userContent, input.apiKey, input.model, maxTokens, temperature);
203+
const result = await callLLM(systemPrompt, userContent, input.apiKey, input.model, maxTokens, temperature, sessionId);
203204

204205
// Strip any markdown code blocks or quotes the LLM might add
205206
const cleaned = result
@@ -220,7 +221,8 @@ Rules:
220221

221222
async function validateTranslation(
222223
input: TranslationRequest,
223-
translatedText: string
224+
translatedText: string,
225+
sessionId?: string
224226
): Promise<{ isValid: boolean; qualityScore: number; issues: string[]; suggestion?: string }> {
225227
const isSameLanguage = input.sourceLanguage === input.targetLanguage ||
226228
(input.sourceLanguage === 'auto' && input.targetLanguage === 'en');
@@ -273,7 +275,7 @@ ${translatedText}
273275
"""`;
274276

275277
// With reasoning_effort "none", 1024 is enough for validation JSON response
276-
const result = await callLLM(systemPrompt, userContent, input.apiKey, input.model, 1024, 0.1);
278+
const result = await callLLM(systemPrompt, userContent, input.apiKey, input.model, 1024, 0.1, sessionId);
277279

278280
try {
279281
const jsonMatch = result.match(/\{[\s\S]*\}/);
@@ -297,7 +299,8 @@ ${translatedText}
297299
async function refineTranslation(
298300
input: TranslationRequest,
299301
translatedText: string,
300-
issues: string[]
302+
issues: string[],
303+
sessionId?: string
301304
): Promise<string> {
302305
const issuesList = issues.map(i => `- ${i}`).join('\n');
303306
const isSameLanguage = input.sourceLanguage === input.targetLanguage ||
@@ -326,7 +329,7 @@ ${translatedText}
326329
Issues found with the current translation:
327330
${issuesList}`;
328331

329-
const result = await callLLM(systemPrompt, userContent, input.apiKey, input.model, calculateMaxTokens(translatedText, isSameLanguage), 0.2);
332+
const result = await callLLM(systemPrompt, userContent, input.apiKey, input.model, calculateMaxTokens(translatedText, isSameLanguage), 0.2, sessionId);
330333

331334
return result
332335
.replace(/^```[\w]*\n?/m, '')
@@ -341,20 +344,21 @@ ${issuesList}`;
341344
export async function runFastTranslation(input: TranslationRequest): Promise<TranslationResult> {
342345
console.log('[Pipeline] Fast mode: translate only (no validate/refine)');
343346

347+
const sessionId = randomUUID();
344348
let translatedText = '';
345349
let model = input.model || DEFAULT_MODEL;
346350
const pipeline: string[] = ['fast-translate'];
347351

348352
try {
349-
const result = await translateText(input);
353+
const result = await translateText(input, false, sessionId);
350354
translatedText = result.translatedText;
351355
model = result.model;
352356

353357
// Echo detection — retry once if model echoed input
354358
if (isEcho(input.text, translatedText) && input.sourceLanguage !== input.targetLanguage) {
355359
console.log('[Pipeline] Echo detected in fast mode — retrying...');
356360
pipeline.push('echo-detected', 'fast-retry');
357-
const retryResult = await translateText(input, true);
361+
const retryResult = await translateText(input, true, sessionId);
358362
translatedText = retryResult.translatedText;
359363
}
360364
} catch (err: unknown) {
@@ -389,6 +393,7 @@ export async function runFastTranslation(input: TranslationRequest): Promise<Tra
389393
export async function runTranslationPipeline(input: TranslationRequest): Promise<TranslationResult> {
390394
console.log(`[Pipeline] Full pipeline: src=${input.sourceLanguage}, target=${input.targetLanguage}, text length=${input.text.length}`);
391395

396+
const sessionId = randomUUID();
392397
let attempt = 0;
393398
let refinements = 0;
394399
const maxRefinements = 2;
@@ -408,7 +413,7 @@ export async function runTranslationPipeline(input: TranslationRequest): Promise
408413
pipeline.push('translate');
409414

410415
try {
411-
const result = await translateText(input);
416+
const result = await translateText(input, false, sessionId);
412417
translatedText = result.translatedText;
413418
model = result.model;
414419

@@ -419,7 +424,7 @@ export async function runTranslationPipeline(input: TranslationRequest): Promise
419424
attempt++;
420425
pipeline.push('translate-retry');
421426

422-
const retryResult = await translateText(input, true);
427+
const retryResult = await translateText(input, true, sessionId);
423428
translatedText = retryResult.translatedText;
424429

425430
// If still echoing after retry, note it but continue to validation
@@ -444,7 +449,7 @@ export async function runTranslationPipeline(input: TranslationRequest): Promise
444449
console.log(`[Pipeline] Translate retry context: ${fixContext.substring(0, 200)}`);
445450

446451
try {
447-
const result = await translateText(input, true);
452+
const result = await translateText(input, true, sessionId);
448453
translatedText = result.translatedText;
449454
model = result.model;
450455

@@ -488,7 +493,7 @@ export async function runTranslationPipeline(input: TranslationRequest): Promise
488493
pipeline.push('validate');
489494

490495
try {
491-
const validation = await validateTranslation(input, translatedText);
496+
const validation = await validateTranslation(input, translatedText, sessionId);
492497
qualityScore = validation.qualityScore;
493498
issues = validation.issues;
494499

@@ -526,12 +531,12 @@ export async function runTranslationPipeline(input: TranslationRequest): Promise
526531
console.log(`[Pipeline] Refinement #${refinements} context: ${fixContext.substring(0, 200)}`);
527532

528533
try {
529-
translatedText = await refineTranslation(input, translatedText, issues);
534+
translatedText = await refineTranslation(input, translatedText, issues, sessionId);
530535

531536
// Re-validate after refinement
532537
pipeline.push(`revalidate-${refinements}`);
533538
try {
534-
const revalidation = await validateTranslation(input, translatedText);
539+
const revalidation = await validateTranslation(input, translatedText, sessionId);
535540
qualityScore = revalidation.qualityScore;
536541
issues = revalidation.issues;
537542

0 commit comments

Comments
 (0)