Skip to content

⭐️⭐️⭐️ Open-source, self-hosted alternative to Linear and Plane. Project management and issue tracking where teams and AI agents work side by side to plan and ship products.

fix(api): send x-opencode-session to OpenCode Go (#440)

Croff4c5c4a67a187
diff --git a/apps/api/src/modules/agents/core/runtime/index.ts b/apps/api/src/modules/agents/core/runtime/index.ts
index 57baccdc..77ebac47 100644
--- a/apps/api/src/modules/agents/core/runtime/index.ts
+++ b/apps/api/src/modules/agents/core/runtime/index.ts
@@ -1,3 +1,4 @@
+import { randomUUID } from 'node:crypto';
import { Agent } from '@mastra/core/agent';
import { getAgentInProject, getInternalAgentApiKey, type AiAgentRow } from '../service';
import { getProjectById } from '#modules/projects/service';
@@ -35,12 +36,19 @@ const DEFAULT_MAX_STEPS = 12;
// Mastra's model config: an object carrying the provider id, model id, and the
// explicit apiKey (and url for OpenAI-compatible endpoints).
-type ModelConfig = { providerId: string; modelId: string; apiKey: string; url?: string };
+type ModelConfig = {
+ providerId: string;
+ modelId: string;
+ apiKey: string;
+ url?: string;
+ headers?: Record<string, string>;
+};
// Builds the model config for an agent from its model credential. The credential's
// integration key is the provider id; its decrypted config carries the apiKey and an
-// optional base URL. The model id is stored on the agent.
-async function resolveModel(row: AiAgentRow): Promise<ModelConfig> {
+// optional base URL. The model id is stored on the agent. OpenCode Go rejects a request
+// without an x-opencode-session header, which has to stay the same across one conversation.
+async function resolveModel(row: AiAgentRow, sessionId: string): Promise<ModelConfig> {
if (row.modelCredentialId == null) {
throw new HttpError(400, 'Agent has no model credential set');
}
@@ -57,6 +65,7 @@ async function resolveModel(row: AiAgentRow): Promise<ModelConfig> {
modelId,
apiKey: String(secret.config.apiKey ?? ''),
...(baseUrl ? { url: baseUrl } : {}),
+ ...(provider === 'opencode-go' ? { headers: { 'x-opencode-session': sessionId } } : {}),
};
}
@@ -75,10 +84,11 @@ async function buildAgent(
row: AiAgentRow,
projectId: number,
contextPreamble: string,
+ sessionId: string,
): Promise<Agent> {
const project = await getProjectById(projectId);
if (!project) throw new HttpError(404, 'Project not found');
- const model = await resolveModel(row);
+ const model = await resolveModel(row, sessionId);
const skills = await listAgentSkills(row.id);
const customTools = await listAgentToolsForRun(row.id);
const apiKey = await getInternalAgentApiKey(row);
@@ -219,21 +229,28 @@ async function prepareRun(
if (row.kind !== 'internal') {
throw new HttpError(400, 'Only internal agents can be run');
}
- const agent = await buildAgent(row, projectId, opts.contextPreamble ?? '');
+ // A chat run with no threadId starts a new conversation; every other run continues
+ // the thread its caller named (see thread-ids). Without memory every run is a
+ // conversation of its own, which is what the model's session id follows.
+ const threadId = row.memoryEnabled
+ ? (opts.threadId ?? newChatThreadId(row.id, opts.callerUserId))
+ : null;
+ const agent = await buildAgent(
+ row,
+ projectId,
+ opts.contextPreamble ?? '',
+ threadId ?? randomUUID(),
+ );
// Mastra's generate/stream have overloaded options; type the shape we use. In
// Mastra v1 the temperature belongs to the call's model settings — a flat
// `temperature` option is only read by the legacy generate.
const options: RunOptions = { maxSteps: row.maxSteps ?? DEFAULT_MAX_STEPS };
if (row.temperature != null) options.modelSettings = { temperature: row.temperature };
- let threadId: string | null = null;
- if (row.memoryEnabled) {
- // A chat run with no threadId starts a new conversation; every other run continues
- // the thread its caller named (see thread-ids). The thread is created up front with
- // what it is bound to and a title, so the chat history can list it and deleting the
- // agent, project or issue can find it. The threadId used is returned so the caller
- // can continue.
- threadId = opts.threadId ?? newChatThreadId(row.id, opts.callerUserId);
+ if (threadId) {
+ // The thread is created up front with what it is bound to and a title, so the chat
+ // history can list it and deleting the agent, project or issue can find it. The
+ // threadId used is returned so the caller can continue.
await ensureThread(
threadId,
opts.callerUserId,