feat(spend-control): evaluation, enforcement, and usage-trigger runtime by subomi · Pull Request #4548 · speakeasy-api/gram

subomi · 2026-07-24T20:10:34Z

Summary by cubic

Ships Budgets v1: a spend-rules API, dashboard, and a runtime that evaluates spend and enforces blocks within seconds. Enforcement is behind the Budgets flag and applies to Claude only.

New Features
- Spend rules API: create/list/get/update/archive, preview matched actors and current-window usage, list warning/breach events, overview metrics, and list actor attributes; OpenAPI, CLI, and generated @gram/client hooks included.
- Dashboard: new Costs → Budgets tab with rules list, detail, editor with live preview, events feed, and overview cards; targets use the backend actor-attribute catalog.
- Runtime: Temporal workflows schedule org sweeps and run debounced per-org evaluations; evaluator computes per-actor window spend from ClickHouse, records warning/breach events, and publishes Redis gate snapshots; telemetry logger observes Claude Code OTEL and Codex/Cursor usage on both server and worker paths and signals re-evaluation with per-org throttling; Claude hooks consult the spend gate before policy scans and block prompts and tool calls (native and MCP, even mid-turn) with a user-facing reason and block URL.
Bug Fixes
- Usage signals run off the write path in a bounded, cancellation-detached context and flush on shutdown.
- Evaluator fails loudly without ClickHouse, retries transient event writes, and logs-and-skips per-rule logic errors.
- Gate rejects nil cache, bounds the compiled-CEL cache, lifts blocks when windows reset, clears snapshots when the flag is off or no rules exist, and uses a longer snapshot TTL (4x the evaluation interval).
- Debounced workflow continues-as-new when an error occurs and a signal arrived mid-run, so pending signals aren’t stranded.
- Dedupes blocked prompt telemetry per session+prompt for clients that lack idempotency keys.
- Key actor evaluation and gate snapshots by user ID (not email) to avoid aliasing across addresses.

^{Written for commit 8952dfb. Summary will update on new commits.}

changeset-bot · 2026-07-24T20:10:39Z

⚠️ No Changeset found

Latest commit: 8952dfb

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

This PR includes no changesets

When changesets are added to this PR, you'll see the packages that this PR includes changesets for and the associated semver types

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

cubic-dev-ai

All reported issues were addressed across 19 files

_{Tip: instead of fixing issues one by one fix them all with cubic

Re-trigger cubic}

Stacked on the spend-control management API PR. Adds the runtime that evaluates and enforces budget rules: - Background evaluator: temporal workflow/activity computing per-actor window spend and opening/closing warning + breach circuits. - Hooks spend gate: blocks over-budget actors on the ingest path. - Telemetry usage trigger: throttled per-org re-evaluation on fresh spend-relevant telemetry, wired via a telemetry LogObserver. - Wires the concrete evaluation signaler into the spend rules API service and registers the worker-side activity + observers. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

- UsageTrigger: fire signals off the telemetry write path in a bounded, cancellation-detached context, so a slow Redis/Temporal dependency can't stall LogBulk, a canceled poll activity still wakes enforcement, and graceful shutdown stays within its drain deadline. - EvaluateOrg: fail loudly instead of panicking when the worker has no ClickHouse connection; surface transient event-write failures so the activity retries (events dedupe on their unique index); log-and-skip per-rule logic errors rather than aborting the org. - Gate: reject a nil cache at construction (fail-open contract), and bound the compiled-CEL cache so repeated rule edits can't grow memory. - Raise the gate snapshot TTL to 4x the evaluation interval so a slow scheduled sweep does not let snapshots expire and enforcement fail open. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

cubic-dev-ai

All reported issues were addressed across 15 files (changes from recent commits).

_{Tip: instead of fixing issues one by one fix them all with cubic

Re-trigger cubic}

cubic-dev-ai

All reported issues were addressed across 9 files (changes from recent commits).

_{Tip: Review your code locally with the cubic CLI to iterate faster.

Fix all with cubic | Re-trigger cubic}

Co-authored-by: Cursor <cursoragent@cursor.com>

cubic-dev-ai

All reported issues were addressed across 8 files (changes from recent commits).

_{Tip: Review your code locally with the cubic CLI to iterate faster.

Fix all with cubic | Re-trigger cubic}

Co-authored-by: cubic-dev-ai[bot] <191113872+cubic-dev-ai[bot]@users.noreply.github.com>

Co-authored-by: Cursor <cursoragent@cursor.com>

gram-bot

Auto-approved: review:bypass label applied by @subomi. Required status checks still gate this merge.

subomi requested a review from a team as a code owner July 24, 2026 20:10

cubic-dev-ai Bot reviewed Jul 24, 2026

View reviewed changes

subomi force-pushed the cursor/spend-control-runtime-92fd branch 2 times, most recently from 26b1225 to 02e06f8 Compare July 24, 2026 22:08

subomi force-pushed the cursor/spend-control-feature-92fd branch from c1e019d to 178dacd Compare July 24, 2026 22:19

subomi requested a review from a team as a code owner July 24, 2026 22:19

Base automatically changed from cursor/spend-control-feature-92fd to main July 25, 2026 13:04

subomi force-pushed the cursor/spend-control-runtime-92fd branch from 02e06f8 to 0da7498 Compare July 25, 2026 15:13

subomi force-pushed the cursor/spend-control-runtime-92fd branch from 0da7498 to 0edd554 Compare July 25, 2026 15:50

subomi force-pushed the cursor/spend-control-runtime-92fd branch from 0edd554 to bcb5853 Compare July 25, 2026 17:01

chore: clean up evaluation runtime

f499244

cubic-dev-ai Bot reviewed Jul 27, 2026

View reviewed changes

chore: resolve comments

7f7d2ad

cubic-dev-ai Bot reviewed Jul 27, 2026

View reviewed changes

Comment thread server/internal/spendrules/impl.go Outdated

fix(spend-control): address remaining evaluation review findings

fb64251

Co-authored-by: Cursor <cursoragent@cursor.com>

This comment has been minimized.

Sign in to view

cubic-dev-ai Bot reviewed Jul 27, 2026

View reviewed changes

Comment thread server/internal/spendrules/impl.go

subomi and others added 2 commits July 27, 2026 16:38

Update server/internal/hooks/claude_hooks.go

0bf64a9

Co-authored-by: cubic-dev-ai[bot] <191113872+cubic-dev-ai[bot]@users.noreply.github.com>

fix(spend-control): key actor evaluation by user id

8952dfb

Co-authored-by: Cursor <cursoragent@cursor.com>

subomi added the review:bypass Merge without human review approval. Required status checks still apply. label Jul 27, 2026

subomi enabled auto-merge July 27, 2026 21:41

gram-bot Bot approved these changes Jul 27, 2026

View reviewed changes

subomi added this pull request to the merge queue Jul 27, 2026

Merged via the queue into main with commit 4c488a4 Jul 27, 2026
42 checks passed

subomi deleted the cursor/spend-control-runtime-92fd branch July 27, 2026 21:49

github-actions Bot locked and limited conversation to collaborators Jul 27, 2026

Uh oh!

Conversation

subomi commented Jul 24, 2026 • edited by cubic-dev-ai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by cubic

Uh oh!

changeset-bot Bot commented Jul 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

⚠️ No Changeset found

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

This comment has been minimized.

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

gram-bot Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

subomi commented Jul 24, 2026 •

edited by cubic-dev-ai Bot

Loading

changeset-bot Bot commented Jul 24, 2026 •

edited

Loading

cubic-dev-ai Bot left a comment •

edited

Loading

cubic-dev-ai Bot left a comment •

edited

Loading

cubic-dev-ai Bot left a comment •

edited

Loading

cubic-dev-ai Bot left a comment •

edited

Loading