- Category
- Developer Tools
- Rank
- No. 1410Tools index
Previous survey · No. 1416 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- getsentry
- GitHub
- 78 stars
- Date
About
Tracks token consumption of coding agents so teams can monitor LLM spend.
What it does
Abacus consolidates coding-tool usage into a team dashboard with cost estimates, model breakdowns, user histories, trends, and commit attribution. It ingests Claude Code data through Anthropic’s admin API, Cursor data through its admin API or CSV, and GitHub commit signals through webhooks and API access. Scheduled synchronization and manual backfills keep the dataset current.
Why it's ranked here
Abacus is a strong choice for teams using Claude Code and Cursor together. It combines provider data, identity mapping, period comparisons, per-user analysis, and AI-attributed commits in one deployable application. The focused provider list and infrastructure requirements make it less compelling for teams with broader tool fleets or different hosting preferences.
What's good
The schema handles both aggregated Anthropic records and event-level Cursor records, with provider-aware deduplication and incremental sync state. Dashboards expose costs, tokens, active users, model preferences, daily trends, and previous-period comparisons. CSV imports cover missing APIs and historical backfills. Google domain enforcement restricts account creation server-side, rather than relying only on the account picker.
Tradeoffs
Built-in usage ingestion covers Claude Code and Cursor only, while commit attribution targets GitHub. Running it requires Postgres, Google OAuth credentials, several secrets, and provider admin keys. The stack is strongly oriented toward Vercel. Client monitoring enables default PII transmission when Sentry is configured, so operators should review that setting before deployment.
How to use it well
Use Abacus as an internal analytics layer for an engineering organization already standardized on Claude Code, Cursor, or both. Schedule provider syncs, import CSV files for gaps, map identities, and review costs alongside user and commit trends. It does not run coding agents, choose models for developers, or enforce purchasing and budget policy.
Technical notes+
The private pnpm package in package.json runs a Next.js 16 and React 19 application with TypeScript, Drizzle ORM, Vercel Postgres, Better Auth, Vitest, and a tsx-based CLI. src/lib/schema.ts defines usage, identity, sync-state, repository, commit, and multi-tool attribution tables, including a composite unique index for provider-specific deduplication. src/lib/queries.ts mixes Drizzle queries with the Vercel SQL template and calculates current-versus-previous-period aggregates. src/proxy.ts protects application and API routes using a Better Auth session cookie while exempting auth, cron, webhook, and sign-in routes. src/lib/auth.ts configures Google OAuth, twelve-hour sessions, and server-side email-domain enforcement. src/instrumentation-client.ts sets full trace sampling, enables logs, and enables default PII for Sentry client telemetry.
Observed
- License
- Apache 2.0
- Primary language
- TypeScript
- Packaging
- Private pnpm package with development, build, start, lint, test, migration, backfill, and CLI scripts
- Interfaces
- Web dashboard, command-line sync and backfill tools, CSV import, provider admin APIs, and GitHub webhooks
- Built-in providers
- Claude Code through Anthropic Admin API, Cursor through Admin API or CSV, and GitHub commit tracking
- Deployment surface
- Vercel deployment is documented, with local development also supported
- Data layer
- Postgres through Drizzle ORM, Vercel Postgres, and Neon serverless pooling
- Authentication
- Google OAuth with server-side email-domain restriction
Read from README.md, package.json, src/proxy.ts, src/instrumentation.ts, src/instrumentation-client.ts, src/lib/db.ts, src/lib/csv.ts, src/lib/auth.ts, src/lib/tips.ts, src/app/page.tsx, src/lib/schema.ts, src/app/layout.tsx, src/lib/queries.ts, src/lib/constants.ts, src/lib/dateUtils.ts.
What it can do
Track token consumption of coding agents
Coding agent activity data → Token usage metrics
Monitor LLM spending
Token usage and pricing data → Cost reports and spending analytics
Generate team spending reports
Team member activities and token usage → Team-level cost breakdown reports
Calculate cost per coding session
Individual coding agent sessions → Per-session cost calculations
Aggregate LLM usage across projects
Multiple project token consumption data → Cross-project usage summaries
Intel on Abacus
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.