Using Firebase Logic AI #3686

Activity

1. JGSolutions commented on Dec 12, 2025

   

   JGSolutions

   on Dec 12, 2025

   Author

   More actions

   here is the js 403 error I get:

   https://firebasevertexai.googleapis.com/v1beta/projects/board-games-315b9/models/gemini-3-pro-preview:streamGenerateContent?alt=sse: [403 ] Requests to this API firebasevertexai.googleapis.com method google.firebase.vertexai.v1beta.GenerativeService.StreamGenerateContent are blocked.

   I know I have to enable the vertex i don't want to. I want to use the Gemini
2. gui-complypro commented on Dec 18, 2025

   

   gui-complypro

   on Dec 18, 2025

   More actions

   Preview models are only available in the global region.

   //https://firebase.google.com/docs/ai-logic/models#available-model-names
   const vertexAI = getAI(firebaseApp, { backend: new VertexAIBackend('global') });
   this.model = getGenerativeModel(vertexAI, {
     model: 'gemini-3-flash-preview',
     systemInstruction: systemPrompt,
   });
3. armando-navarro

   added a commit that references this issue on Jul 27, 2026

   fix(schematics): point the ng add AI feature at @angular/fire/ai (#3685)

   ...

   3e671e0
4. armando-navarro

   added

   type: featureNew capability or enhancement.New capability or enhancement.

   comp: aiFirebase AI Logic / Vertex AI (src/ai).Firebase AI Logic / Vertex AI (src/ai).

   version: current (v17+)Targets the current modular API (v17+).Targets the current modular API (v17+).

   triaged: keep-openTriaged and intentionally kept open; exempt from any future stale process.Triaged and intentionally kept open; exempt from any future stale process.

   on Aug 3, 2026
5. armando-navarro

   added a commit that references this issue on Aug 3, 2026

   feat(schematics): rewrite Vertex AI imports to AI Logic on ng update (#…

   ...

   b551b5f
6. KingDarBoja commented on Oct 27, 2026

   

   KingDarBoja

   on Oct 27, 2026

   Contributor

   More actions

   This should be fixed already with latest AngularFire version (v21).
7. armando-navarro commented on Nov 5, 2026

   

   armando-navarro

   on Nov 5, 2026

   Collaborator

   More actions

   Thanks for the report, and thanks @gui-complypro for the note about preview models.

   To answer the question in the issue: yes, GoogleAIBackend selects the Gemini Developer API.

   • In the Firebase JS SDK both backends send their requests to the firebasevertexai.googleapis.com host, and the backend only changes the path.
   • The Gemini Developer API path is projects/{project}/models/{model}, which is the path in your error. The Vertex AI path adds a locations/{location} segment.
   • AngularFire passes your getAI options to the SDK unchanged, so the backend you asked for is the one in effect.

   That makes the 403 a project setting rather than a backend mix-up. Firebase's AI Logic error-codes page lists this exact message, "Requests to this API firebasevertexai.googleapis.com ... are blocked", under API key restrictions:

   • The API key in your app's Firebase config excludes the Firebase AI Logic API.
   • The fix is in the Google Cloud console under APIs & Services, Credentials: add the Firebase AI Logic API to that key's API restrictions. It can take a few minutes to apply.
   • The Gemini Developer API does not need Vertex AI and runs on the Spark plan. Preview models such as gemini-3-pro-preview need the Blaze plan, whichever provider you use.

   AngularFire 21.0.0 ships @angular/fire/ai with provideAI, getAI and both backends, and the AI Logic guide covers the setup: https://github.com/angular/angularfire/blob/21.0.x/docs/ai.md

   I'm closing this since there is nothing for AngularFire to change here. If fixing the key restriction does not clear the 403, please open a new issue with the request URL and the key settings, and I'll take another look.
8. armando-navarro

   closed this as not plannedon Nov 5, 2026
9. armando-navarro

   removed

   triaged: keep-openTriaged and intentionally kept open; exempt from any future stale process.Triaged and intentionally kept open; exempt from any future stale process.

   on Nov 5, 2026