feat(subscriptions): gate AI summaries on AI credit budget

[vdekrijger]

Problem

Subscription AI summaries call an LLM on every delivery with no budget check. Now that subscriptions are available on the free tier (#59624), an org could keep generating summaries while over its AI credit budget. We should stop generating once an org is over budget — the same way the chat assistant already behaves.

Changes

Adopt the chat assistant's enforcement pattern (ee/api/conversation.py): is_team_limited(QuotaResource.AI_CREDITS), which reads the billing quota-limiting cache (not the retrospective usage query).

• Enable-time gate (SubscriptionSerializer.validate): raises QuotaLimitExceeded (402) when a summary is being toggled on while the org is over budget. Reuses the existing "becoming active" check, so grandfathered summaries and unrelated edits are unaffected.
• Generation-time gate (snapshot_subscription_insights): when over budget, skips the summary and delivers the rest of the subscription (graceful degradation) rather than failing the delivery. Fail-open — if the quota lookup itself errors, the summary is generated rather than silently dropped.

How did you test this code?

I'm an agent (Claude Code) — automated suites I ran locally:

• ee/api/test/test_subscription.py — enable-time gate over/under budget, the patch-transition case, and the grandfathering case (unrelated edits while over budget aren't blocked).
• posthog/temporal/subscriptions/test_snapshot_ai_consent.py — generation-time skip when over budget, generate when under, and fail-open when the quota check raises.

Publish to changelog?

No — internal cost-control guard, no user-facing feature.

Docs update

n/a

🤖 Agent context

Authored with Claude Code (human-directed). Stacked on #59624.

Investigation that shaped this: the AI credit budget shown in /usage is computed by a retrospective ClickHouse query and is display-only — the actual enforcement signal is is_team_limited(AI_CREDITS), populated in a Redis set by the billing quota-limiting system and (until this PR) used only by the Max chat assistant. We reuse that chokepoint rather than introducing a bespoke check, so this is the second adopter of an established pattern, not new billing infrastructure. The lookup is cheap (in-process-cached set membership), so it's safe in the per-delivery path. Behaviour differs by context deliberately: a hard 402 at enable-time (API), graceful skip-and-deliver at generation-time (background workflow).

[vdekrijger]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• feat(subscriptions): gate AI subscriptions on AI credit budget #59634
• feat(subscriptions): expose AI subscriptions + ad-hoc reports as MCP tools #59633
• feat(subscriptions): AI subscription API, validation, telemetry, and UI #59632
• feat(subscriptions): scheduled AI delivery via Temporal (email + Slack) #59631
• feat(subscriptions): AI report pipeline primitive (planner, HogQL, synthesis) + sanitization #59630
• feat(subscriptions): add AI-prompt subscription fields + activity logging #59629
• feat(subscriptions): gate AI summaries on AI credit budget #59625 👈 (View in Graphite)
• feat(subscriptions): bring subscriptions to the free tier (freemium, synced to alerts) #59624
• master

This stack of pull requests is managed by Graphite. Learn more about stacking.

[github-actions]

🎭 Playwright didn't run on this PR — your changes touch code that could affect E2E behavior, but Playwright is opt-in via label now to keep CI cost down.

Add the run-playwright label if you want an E2E sweep before merging — CI will pick it up automatically.

Most PRs don't need this. Real regressions still get caught on master and fix-forward.

[github-actions]

ClickHouse migration SQL per cloud environment

No ClickHouse migrations changed in this PR.

[greptile-apps]

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
ee/api/subscription.py:289-292
The `get_organization()` call (which may execute a DB query) happens before `_validate_summary_credit_budget()`, the cheaper Redis/cache check. If the budget check raises `QuotaLimitExceeded`, the org fetch was wasted. Placing the fast guard first avoids unnecessary work.

```suggestion
        if self._is_becoming_active_summary(attrs):
            self._validate_summary_credit_budget()
            organization = self.context["get_organization"]()
            self._validate_summary_enabled_org_limit(organization)
```

_{Reviews (1): Last reviewed commit: "feat(subscriptions): gate AI summaries o..." | Re-trigger Greptile}

[greptile-apps]

_{Reviews (2): Last reviewed commit: "feat(subscriptions): notify when AI summ..." | Re-trigger Greptile}

[greptile-apps]

Want your agent to iterate on Greptile's feedback? Try greploops.

[greptile-apps]

_{Reviews (3): Last reviewed commit: "chore(subscriptions): trim verbose comme..." | Re-trigger Greptile}