commit 38b2bf6eb66ba1d401d211873670ff4c8ae59059
parent dc4e9d14937fb0fff9c2cc427826df18002bd4fb
Author: Christopher DiPersio <cdipersio@mozilla.com>
Date: Tue, 30 Dec 2025 21:20:43 +0000
Bug 2006939 - Add required headers to OpenAIEngine class r=tzhang,npodgurski,ai-models-reviewers,ai-ondevice-reviewers,gregtatum
Differential Revision: https://phabricator.services.mozilla.com/D277215
Diffstat:
8 files changed, 131 insertions(+), 36 deletions(-)
diff --git a/browser/components/aiwindow/models/Chat.sys.mjs b/browser/components/aiwindow/models/Chat.sys.mjs
@@ -4,14 +4,8 @@
* file, You can obtain one at http://mozilla.org/MPL/2.0/.
*/
-/* eslint-disable-next-line mozilla/reject-import-system-module-from-non-system */
-import { getFxAccountsSingleton } from "resource://gre/modules/FxAccounts.sys.mjs";
import { openAIEngine } from "moz-src:///browser/components/aiwindow/models/Utils.sys.mjs";
import {
- OAUTH_CLIENT_ID,
- SCOPE_PROFILE,
-} from "resource://gre/modules/FxAccountsCommon.sys.mjs";
-import {
toolsConfig,
getOpenTabs,
searchBrowsingHistory,
@@ -28,21 +22,6 @@ export const Chat = {
get_page_content: GetPageContent.getPageContent.bind(GetPageContent),
},
- async _getFxAccountToken() {
- try {
- const fxAccounts = getFxAccountsSingleton();
- const token = await fxAccounts.getOAuthToken({
- // Scope needs to be updated in accordance with https://bugzilla.mozilla.org/show_bug.cgi?id=2005290
- scope: SCOPE_PROFILE,
- client_id: OAUTH_CLIENT_ID,
- });
- return token;
- } catch (error) {
- console.warn("Error obtaining FxA token:", error);
- return null;
- }
- },
-
/**
* Stream assistant output with tool-call support.
* Yields assistant text chunks as they arrive. If the model issues tool calls,
@@ -56,7 +35,7 @@ export const Chat = {
const engineInstance = await openAIEngine.build();
// Note FXA token fetching disabled for now - this is still in progress
// We can flip this switch on when more realiable
- const fxAccountToken = await this._getFxAccountToken();
+ const fxAccountToken = await openAIEngine.getFxAccountToken();
// We'll mutate a local copy of the thread as we loop
// We also filter out empty assistant messages because
diff --git a/browser/components/aiwindow/models/Insights.sys.mjs b/browser/components/aiwindow/models/Insights.sys.mjs
@@ -24,7 +24,7 @@
*
*/
-import { renderPrompt } from "./Utils.sys.mjs";
+import { renderPrompt, openAIEngine } from "./Utils.sys.mjs";
import {
HISTORY,
@@ -365,6 +365,7 @@ function normalizeInsightList(parsed) {
*/
export async function generateInitialInsightsList(engine, sources) {
const promptText = await buildInitialInsightsGenerationPrompt(sources);
+
const response = await engine.run({
args: [
{
@@ -374,6 +375,7 @@ export async function generateInitialInsightsList(engine, sources) {
{ role: "user", content: promptText },
],
responseFormat: { type: "json_schema", schema: INITIAL_INSIGHTS_SCHEMA },
+ fxAccountToken: await openAIEngine.getFxAccountToken(),
});
const parsed = parseAndExtractJSON(response, []);
@@ -397,6 +399,7 @@ export async function deduplicateInsights(
existingInsightsList,
newInsightsList
);
+
const response = await engine.run({
args: [
{
@@ -409,6 +412,7 @@ export async function deduplicateInsights(
type: "json_schema",
schema: INSIGHTS_DEDUPLICATION_SCHEMA,
},
+ fxAccountToken: await openAIEngine.getFxAccountToken(),
});
const parsed = parseAndExtractJSON(response, { unique_insights: [] });
@@ -440,6 +444,7 @@ export async function deduplicateInsights(
export async function filterSensitiveInsights(engine, insightsList) {
const sensitivityFilterPrompt =
await buildInsightsSensitivityFilterPrompt(insightsList);
+
const response = await engine.run({
args: [
{
@@ -452,6 +457,7 @@ export async function filterSensitiveInsights(engine, insightsList) {
type: "json_schema",
schema: INSIGHTS_NON_SENSITIVE_SCHEMA,
},
+ fxAccountToken: await openAIEngine.getFxAccountToken(),
});
const parsed = parseAndExtractJSON(response, { non_sensitive_insights: [] });
diff --git a/browser/components/aiwindow/models/InsightsManager.sys.mjs b/browser/components/aiwindow/models/InsightsManager.sys.mjs
@@ -429,6 +429,7 @@ export class InsightsManager {
type: "json_schema",
schema: INSIGHTS_MESSAGE_CLASSIFY_SCHEMA,
},
+ fxAccountToken: await openAIEngine.getFxAccountToken(),
});
const parsed = parseAndExtractJSON(response, {
diff --git a/browser/components/aiwindow/models/TitleGeneration.sys.mjs b/browser/components/aiwindow/models/TitleGeneration.sys.mjs
@@ -60,7 +60,10 @@ export async function generateChatTitle(message, current_tab) {
];
// Call the LLM
- const response = await engine.run({ messages });
+ const response = await engine.run({
+ messages,
+ fxAccountToken: await openAIEngine.getFxAccountToken(),
+ });
// Extract the generated title from the response
const title =
diff --git a/browser/components/aiwindow/models/Utils.sys.mjs b/browser/components/aiwindow/models/Utils.sys.mjs
@@ -11,6 +11,11 @@
*/
import { createEngine } from "chrome://global/content/ml/EngineProcess.sys.mjs";
+import { getFxAccountsSingleton } from "resource://gre/modules/FxAccounts.sys.mjs";
+import {
+ OAUTH_CLIENT_ID,
+ SCOPE_PROFILE,
+} from "resource://gre/modules/FxAccountsCommon.sys.mjs";
/**
* openAIEngine class
@@ -23,19 +28,61 @@ export class openAIEngine {
*/
static _createEngine = createEngine;
- static async build(engineId = "smart-openai") {
+ /**
+ * Returns an OpenAIEngine instance with the specified engine and service types
+ *
+ * @param {string} engineId The identifier for the engine instance
+ * @param {string} serviceType The type of message to be sent ("ai", "memories", "s2s")
+ * @returns {Promise<openAIEngine>} The OpenAIEngine instance
+ */
+ static async build(engineId = "smart-openai", serviceType = "ai") {
const engine = new openAIEngine();
- engine.engineInstance = await openAIEngine.#createOpenAIEngine(engineId);
+ engine.engineInstance = await openAIEngine.#createOpenAIEngine(
+ engineId,
+ serviceType
+ );
return engine;
}
/**
+ * Retrieves the Firefox account token
+ *
+ * @returns {Promise<string|null>} The Firefox account token (string) or null
+ */
+ static async getFxAccountToken() {
+ try {
+ const fxAccounts = getFxAccountsSingleton();
+ return await fxAccounts.getOAuthToken({
+ // Scope needs to be updated in accordance with https://bugzilla.mozilla.org/show_bug.cgi?id=2005290
+ scope: SCOPE_PROFILE,
+ client_id: OAUTH_CLIENT_ID,
+ });
+ } catch (error) {
+ console.warn("Error obtaining FxA token:", error);
+ return null;
+ }
+ }
+
+ /**
* Creates an OpenAI engine instance
*
- * @param {string} engineId The identifier for the engine instance
- * @returns {Promise<object>} The configured engine instance
+ * @param {string} engineId The identifier for the engine instance
+ * @param {string} serviceType The type of message to be sent ("ai", "memories", "s2s")
+ * @returns {Promise<object>} The configured engine instance
*/
- static async #createOpenAIEngine(engineId) {
+ static async #createOpenAIEngine(engineId, serviceType) {
+ const extraHeadersPref = Services.prefs.getStringPref(
+ "browser.aiwindow.extraHeaders",
+ "{}"
+ );
+ let extraHeaders = {};
+ try {
+ extraHeaders = JSON.parse(extraHeadersPref);
+ } catch (e) {
+ console.error("Failed to parse extra headers from prefs:", e);
+ Services.prefs.clearUserPref("browser.aiwindow.extraHeaders");
+ }
+
try {
const engineInstance = await openAIEngine._createEngine({
apiKey: Services.prefs.getStringPref("browser.aiwindow.apiKey"),
@@ -45,6 +92,8 @@ export class openAIEngine {
modelId: Services.prefs.getStringPref("browser.aiwindow.model"),
modelRevision: "main",
taskName: "text-generation",
+ serviceType,
+ extraHeaders,
});
return engineInstance;
} catch (error) {
diff --git a/browser/components/aiwindow/models/tests/xpcshell/test_Utils.js b/browser/components/aiwindow/models/tests/xpcshell/test_Utils.js
@@ -16,10 +16,12 @@ const { sinon } = ChromeUtils.importESModule(
const PREF_API_KEY = "browser.aiwindow.apiKey";
const PREF_ENDPOINT = "browser.aiwindow.endpoint";
const PREF_MODEL = "browser.aiwindow.model";
+const PREF_EXTRA_HEADERS = "browser.aiwindow.extraHeaders";
const API_KEY = "fake-key";
const ENDPOINT = "https://api.fake-endpoint.com/v1";
const MODEL = "fake-model";
+const EXTRA_HEADERS = '{"x-fastly-request": "fake-key"}';
/**
* Cleans up preferences after testing
@@ -39,6 +41,7 @@ add_task(async function test_createOpenAIEngine() {
Services.prefs.setStringPref(PREF_API_KEY, API_KEY);
Services.prefs.setStringPref(PREF_ENDPOINT, ENDPOINT);
Services.prefs.setStringPref(PREF_MODEL, MODEL);
+ Services.prefs.setStringPref(PREF_EXTRA_HEADERS, EXTRA_HEADERS);
const sb = sinon.createSandbox();
try {
@@ -75,6 +78,12 @@ add_task(async function test_createOpenAIEngine() {
"text-generation",
"taskName should be text-generation"
);
+ Assert.equal(opts.serviceType, "ai", "serviceType should be ai");
+ Assert.deepEqual(
+ opts.extraHeaders,
+ JSON.parse(EXTRA_HEADERS),
+ "extraHeaders should come from pref"
+ );
} finally {
sb.restore();
}
diff --git a/toolkit/components/ml/content/EngineProcess.sys.mjs b/toolkit/components/ml/content/EngineProcess.sys.mjs
@@ -596,6 +596,21 @@ export class PipelineOptions {
staticEmbeddingsOptions = null;
/**
+ * The service type for an OpenAIPipeline.
+ *
+ * @type {"ai" | "memories" | "s2s" | null}
+ */
+ serviceType = null;
+
+ /**
+ * This option allows for extra headers to be passed to
+ * OpenAI-API-compatable endpoints
+ *
+ * @type {?Record<string, string>}
+ */
+ extraHeaders = null;
+
+ /**
* Create a PipelineOptions instance.
*
* @param {object} options - The options for the pipeline. Must include mandatory fields.
@@ -801,6 +816,8 @@ export class PipelineOptions {
"baseURL",
"apiKey",
"staticEmbeddingsOptions",
+ "serviceType",
+ "extraHeaders",
];
if (options instanceof PipelineOptions) {
@@ -946,6 +963,8 @@ export class PipelineOptions {
baseURL: this.baseURL,
apiKey: this.apiKey,
staticEmbeddingsOptions: this.staticEmbeddingsOptions,
+ serviceType: this.serviceType,
+ extraHeaders: this.extraHeaders,
};
}
diff --git a/toolkit/components/ml/content/backends/OpenAIPipeline.mjs b/toolkit/components/ml/content/backends/OpenAIPipeline.mjs
@@ -326,20 +326,49 @@ export class OpenAIPipeline {
) {
lazy.console.debug("Running OpenAI pipeline");
try {
- const { baseURL, apiKey, modelId } = this.#options;
+ const { baseURL, apiKey, modelId, serviceType, extraHeaders, engineId } =
+ this.#options;
const fxAccountToken = request.fxAccountToken
? request.fxAccountToken
: null;
- const defaultHeaders = fxAccountToken
- ? {
- Authorization: `Bearer ${fxAccountToken}`,
- "service-type": "ai",
+
+ let isFastlyRequest = false;
+ if (extraHeaders) {
+ for (const headerKey of Object.keys(extraHeaders)) {
+ if (headerKey.toLowerCase() == "x-fastly-request") {
+ isFastlyRequest = true;
+ break;
}
- : undefined;
+ }
+ }
+
+ /** @type {Record<string, string>} */
+ let authHeaders;
+ if (isFastlyRequest) {
+ // If the x-fastly-request extra header is present, we want to hit the LiteLLM
+ // endpoint directly, so don't use an FxA token
+ authHeaders = {
+ authorization: `Bearer ${apiKey}`,
+ };
+ } else if (fxAccountToken) {
+ // Use a Firefox account token if available
+ authHeaders = {
+ authorization: `Bearer ${fxAccountToken}`,
+ "service-type": serviceType || "ai",
+ };
+ } else {
+ // Don't use any authentication headers
+ authHeaders = {};
+ }
+
const client = new OpenAIPipeline.OpenAILib.OpenAI({
baseURL: baseURL ? baseURL : "http://localhost:11434/v1",
apiKey: apiKey || "ollama",
- ...(defaultHeaders ? { defaultHeaders } : {}),
+ defaultHeaders: {
+ ...authHeaders,
+ ...extraHeaders,
+ "x-engine-id": engineId,
+ },
});
const stream = request.streamOptions?.enabled || false;
const tools = request.tools || [];
