# Perspective API migration guide

Perspective API shuts down on December 31, 2026. Spamgone answers Perspective's `comments:analyze` requests in the same shape, so most integrations switch by changing two things. Using Coral for comments? Follow [Switch Coral from Perspective](/docs/coral) instead.

Switching from Perspective? Use code `PERSPECTIVE50` at checkout for 50% off Pro or Scale for your first 3 months, until January 31, 2027.

## 1. Change the URL and the key

```diff
- https://commentanalyzer.googleapis.com/v1alpha1/comments:analyze?key=GOOGLE_KEY
+ https://spamgone.dev/v1alpha1/comments:analyze?key=SPAMGONE_KEY
```

Google's own client libraries work when you point them at the new base URL. With Google's Python client, for example, pass `client_options={"api_endpoint": "https://spamgone.dev/"}` and your Spamgone key as `developerKey`.

The request and response stay the same: `comment.text`, `requestedAttributes`, `languages`, `spanAnnotations`, `doNotStore` and `clientToken` all work as before.

## 2. Check the attributes you use

| Perspective attribute                        | Spamgone                                                      |
| -------------------------------------------- | ------------------------------------------------------------- |
| `TOXICITY`                                   | Supported: the overall toxicity score.                        |
| `SEVERE_TOXICITY`                            | Supported.                                                    |
| `IDENTITY_ATTACK`                            | Supported.                                                    |
| `INSULT`                                     | Supported.                                                    |
| `PROFANITY`                                  | Supported.                                                    |
| `THREAT`                                     | Supported.                                                    |
| `SEXUALLY_EXPLICIT`                          | Supported: the sexual-content check's `explicit` label.       |
| `SPAM`                                       | Supported, and much stronger: Spamgone's main check.          |
| Experimental attributes such as `FLIRTATION` | Not supported: the request fails with `400 INVALID_ARGUMENT`. |

`comments:suggestscore` saves a suggested `SPAM` score as feedback on the latest check of that text. Suggested scores for other attributes, such as `TOXICITY`, are accepted but not saved, because feedback in Spamgone is about spam.

## 3. Start with these thresholds

Scores are probabilities from 0 to 1, like Perspective's, but they come from a different model, so the same cut-off won't mean the same thing. Good starting points:

- Hide automatically at **0.8** and above.
- Send to human review from **0.5** to 0.8.
- Allow below 0.5.

Run a sample of your own comments through both before moving all traffic, and adjust.

## Known differences

- Scores differ from Perspective's for the same text. Compare on your own data.
- `detectedLanguages` echoes the `languages` you send (or `en`); Spamgone doesn't detect languages.
- Span scores cover the whole comment as one span.
- Errors use Google's `{ "error": { "code", "message", "status" } }` shape, with Spamgone's own messages.
- Quotas and rate limits follow your Spamgone plan, not Perspective's 1 query per second.

## Checklist

- [ ] Create a Spamgone test key and run your existing client against it.
- [ ] Remove any attributes Spamgone doesn't support from `requestedAttributes`.
- [ ] Compare scores on a sample of your comments and set your thresholds.
- [ ] Swap in a live key and set `doNotStore` where you used it before.
- [ ] Watch the first day in the dashboard's Metrics and Review pages.
