canonical: https://jentic.com/apis/googleapis.com/firebaseml # Google Firebase ML API The Firebase ML API provides programmatic access to long-running operations spawned by Firebase ML model management workflows, such as uploading and publishing custom machine learning models for use by Firebase mobile apps. The v1 surface exposes operations to list, cancel, and delete those operations. Use it to monitor model upload pipelines and manage cleanup of operation history during model publishing automation. ## For AI agents Track and clean up long-running Firebase ML model management operations so an agent can supervise custom model upload pipelines. ## Scope Does not train models, perform on-device inference, or upload model artifacts directly via this v1 spec - use for managing Firebase ML long-running operations only. ## Capabilities - List long-running Firebase ML operations under a project - Cancel a Firebase ML custom model upload or publish operation - Delete a completed Firebase ML operation resource - Monitor state of asynchronous Firebase ML model jobs from a pipeline - Reconcile Firebase ML operation history during model release reviews ## Use cases ### Custom Model Upload Monitoring When a Firebase ML pipeline uploads a TensorFlow Lite model to Firebase for distribution to mobile clients, the upload runs as a long-running operation. The Firebase ML v1 API lets release tooling list and inspect those operations to confirm a model has finished processing before promoting it to production via the Firebase ML SDK. Example prompt: Call GET /v1/{+name} on a Firebase ML operation resource and verify done=true with no error before announcing the new model version. ### Cancelling a Stuck Model Operation If a Firebase ML model upload or publish operation hangs because of an invalid model artifact or expired credential, POST /v1/{+name}:cancel on the operation resource cancels the job. This frees the resource so the pipeline can retry the upload with a corrected artifact. Example prompt: Call POST /v1/{+name}:cancel on the in-flight Firebase ML operation and confirm via GET /v1/{+name} that done=true with cancellation reported in the status. ### Firebase ML Operation Hygiene Old Firebase ML operation resources accumulate over time as models are repeatedly uploaded and updated. DELETE /v1/{+name} removes a completed operation record so monitoring views remain focused on current activity. Run as a low-frequency maintenance task alongside model release reviews. Example prompt: Call DELETE /v1/{+name} on a Firebase ML operation that has been done for over 60 days to keep operation listings focused on recent activity. ### Agent-Driven Custom Model Release via Jentic An AI agent connected through Jentic can supervise the upload-and-publish lifecycle of a Firebase ML custom model on behalf of an ML engineer, polling operations to confirm successful processing and cancelling failed jobs. Jentic exposes the 3 Firebase ML v1 endpoints, the agent strings them into a watch-cancel-cleanup workflow, and OAuth 2.0 credentials stay inside your Jentic One instance. Example prompt: Through Jentic, search for monitor firebase ml operations, load the operations.list endpoint, and execute it under projects/PROJECT to flag any operations whose done is still false after 30 minutes. ## Key endpoints | Method | Path | Description | | --- | --- | --- | | GET | /v1/{+name} | List Firebase ML operations under a project | | POST | /v1/{+name}:cancel | Cancel a Firebase ML long-running operation | | DELETE | /v1/{+name} | Delete a Firebase ML operation resource | ## Key resources - **Operations** — List, cancel, and delete Firebase ML long-running operations spawned by model upload and publish workflows ## Why Jentic - **Setup:** Wiring the Firebase ML API by hand means setting up Google OAuth 2.0 with a service account and polling the long-running operation resources to track, cancel, and clean them up. Through Jentic you install once, import the Firebase ML API from the API Directory, store the service-account credential once, and your agent calls it. - **Permission scoping:** This v1 surface acts on operation resources named in the URL path (/v1/{name}, /v1/{name}:cancel), so a rule can pin your agent to one operation and nothing else. You choose the operations it may call, so ones like cancel or delete are not included unless you add them. - **Credential handling:** Your Google service-account credential is stored once, encrypted, by your own Jentic One instance and injected at execution time. It never enters the agent's prompt, logs, or context. - **Discovery method:** Agents search Jentic by intent such as 'monitor a Firebase ML operation' or 'cancel a pending operation', and Jentic returns the matching operation with its input schema so the agent calls the right endpoint without browsing the reference docs. ## Related APIs - **Firebase Management API** — Provisions the Firebase project that owns Firebase ML custom models. - **Google Cloud AI Platform Training and Prediction API** — Hosts and serves trained models from Google Cloud rather than distributing them to mobile devices. - **Google Cloud AutoML API** — Trains custom models without code rather than only managing their distribution. ## FAQ ### What authentication does the Firebase ML API use? The API uses Google OAuth 2.0 with the firebase or cloud-platform scope. Through Jentic the OAuth credentials are stored in the encrypted vault and a scoped access token is supplied per request without exposing the underlying service-account key. ### Can I run model inference with the v1 Firebase ML API endpoints exposed here? No. The v1 surface in this spec covers long-running operation management only. Inference on Firebase ML custom models happens on-device through the Firebase ML SDKs after a model has been uploaded and published, not through this REST API. ### What are the rate limits for the Firebase ML API? Google enforces standard googleapis.com per-project quotas for the Firebase ML management endpoints. Operation listing and cancel are low-cost calls; pipelines that poll frequently should use exponential backoff to stay within per-minute quotas. ### How do I monitor a Firebase ML model upload operation through Jentic? Run pip install jentic, search for monitor firebase ml operations, load the operations.list endpoint, and execute it under projects/PROJECT to find operations whose done is false. Get started with Jentic One, the self-hosted execution layer. ### Does the Firebase ML API support cancelling a stuck custom model upload? Yes. POST /v1/{+name}:cancel on the operation resource cancels the in-flight upload or publish job and reports a cancelled status when complete, letting release pipelines retry with a fixed model artifact. ### Is the Firebase ML API free? Firebase ML custom model hosting is included with Firebase under standard limits, and the operations endpoints themselves do not incur a separate per-call charge. Model storage and download bandwidth fall under standard Firebase project usage. ### Can I limit what my agent is allowed to do with the Firebase ML API? Yes. Because you run Jentic One yourself, your own rules decide which of the three Firebase ML v1 operations the agent may call, so read-only listing at GET /v1/{name} can be permitted while POST /v1/{name}:cancel and DELETE /v1/{name} stay excluded unless you add them. Since each call acts on an operation resource named in the URL path, a rule can pin the agent to one specific operation and nothing else. The Google OAuth 2.0 credential is stored by your instance and injected only when an allowed operation runs.