Add Antigravity local spend and usage history

[robinebers]

TL;DR

Add Antigravity usage history, local token counts, and estimated API-equivalent spend to Today, Yesterday, Last 30 Days, and Total Spend. This is the reviewed, hardened replacement for #1058 and preserves @FelixIsaac as the feature commit's Git author.

Fixes #1120.

What was happening

• Antigravity displayed its Session, Weekly, Claude, and Claude Weekly quotas but never contributed to Total Spend because its token-accounting data was not being scanned.
• Antigravity's JSONL transcripts contain no token counts; the usable generation records live in local SQLite conversation databases as protobuf blobs.
• The original contribution correctly identified that data source, but its all-rows SQLite query could expand enormous blobs into memory, its decoder could trap on malformed values, and its pricing overrides conflicted with existing catalog rates.

What this changes

• Read generation records from ~/.gemini/antigravity-cli/conversations/*.db in bounded SQLite batches and decode only the protobuf fields needed for model names, timestamps, billable system-prompt and input tokens, output tokens, and cached tokens.
• Keep previously decoded conversation history directly on the scanner actor and read only newly appended generation rows on refresh, including changes written to SQLite WAL files.
• Add Antigravity's Usage Trend, Today, Yesterday, and Last 30 Days metrics, export machine-local usage history, and include Antigravity in the Total Spend card.
• Enable the new metrics by default, keep Usage Trend above the fold, place spend rows On Demand, and preserve existing menu-bar pins.
• Resolve Antigravity-specific Gemini model variants through the shared pricing supplement without overriding established model rates or inventing prices for unknown models.
• Warn once for unreadable databases or oversized generation blobs, reject malformed protobuf lengths/counts safely, and document the new provider behavior.
• Preserve Felix Isaac Lim (@FelixIsaac) as the author of the feature commit and explicitly credit the original contribution in the provider documentation.

Heads-up

• Antigravity spend is an estimated API-equivalent value derived from measured local token counts; it is not a charge reported by the user's subscription.
• Generation blobs larger than 1 MiB are skipped with a warning so unusually large conversation snapshots cannot exhaust memory.
• This replaces closed PR feat(antigravity): add transcript log scanner and cost estimation #1058, whose source branch was deleted.

Tests

• swift test -Xlinker -rpath -Xlinker /Users/rebers/.codex/worktrees/7bc2/openusage/.build/out/Products/Debug: 1,331 app tests, four CLI tests, and three Swift Testing cases passed; three existing tests were skipped.
• Added focused coverage for bounded SQL pagination, incremental database and WAL updates, expired-cache eviction, cancellation without partial history, billable system prompts, oversized blobs, malformed protobufs, missing/unpriced models, Gemini aliases, default layout migration, machine-local history classification, and Total Spend aggregation.
• Validated the bounded read-only SQLite query against a real local Antigravity conversation database without exposing conversation contents.

Screenshots

Before: quota meters only

After: quota meters, Usage Trend, and local spend history

Thanks to @FelixIsaac for identifying the conversation-database source, reverse-engineering the generation records, and contributing the original implementation in #1058.

---

Note

Medium Risk
Reads local SQLite conversation DBs and undocumented protobufs, then feeds estimated costs into Total Spend. Bounded queries and skip-on-oversize reduce memory risk, but decoder/pricing mistakes would misreport spend.

Overview
Adds local token usage and estimated API-equivalent spend for Antigravity (Usage Trend, Today, Yesterday, Last 30 Days, and Total Spend). Quota meters stay as they are; spend comes from ~/.gemini/antigravity-cli/conversations/*.db, not JSONL transcripts.

AntigravityDbUsageScanner pages gen_metadata in small SQLite batches, skips blobs over 1 MiB before hex expansion, and incrementally reuses decoded history (including WAL fingerprinting). AntigravityProtoDecoder pulls model, system-prompt + input, output, cache-read, and timestamp fields, treating missing/unpriced models as unknown instead of inventing rates.

Gemini slug aliases in the pricing supplement now cover Antigravity variants (none/a|b|c/preview) without new rate overrides. Default layout enables the new metrics, keeps trend above the fold, and tucks spend rows On Demand.

^{Reviewed by Cursor Bugbot for commit 2fd4fb0. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 9118579. Configure here.}