Compute Atlas

Open-source GPU price aggregation for comparing live cloud and marketplace rates with explicit source coverage and normalization logic.

GitHub repository

Why this exists

GPU pricing is fragmented across cloud vendors and specialized marketplaces. Compute Atlas exists to make the aggregation and normalization path inspectable in public:

• provider inputs are modular and traceable
• methodology is explicit and versioned
• normalized time-series data is queryable
• source gaps remain visible instead of being hand-waved away

This repository is intended to be the auditable core of a practical GPU price tracker. Paid products, if added later, should be service layers on top of the same open codebase rather than a closed replacement for it.

Preview

License

This repository is licensed under Apache-2.0. That keeps the core open and verifiable while still allowing commercial hosted products, support, and other paid service layers to be offered later.

Architecture

• apps/web: Next.js App Router dashboard plus public API
• apps/worker: ingestion, normalization, and optional downstream jobs
• packages/db: schema, migrations, seed data, and query helpers
• packages/provider-sdk: shared adapter interfaces and registry
• packages/providers: live provider implementations
• packages/methodology: normalization and index calculation
• packages/publisher-solana: optional Solana devnet publisher kept separate
• packages/api-contract: typed response schemas and OpenAPI helpers

Open-source model

The repository is open source first.

Open and auditable:

• methodology logic
• provider adapter interfaces
• public API contracts
• dashboard transparency views

Possible paid layers later:

• managed hosted dashboard
• alerting and notifications
• higher-availability endpoints
• premium historical analytics
• team and access-control features

Provider source reality

As of March 25, 2026, provider coverage is mixed:

• strong official API paths: AWS, Azure, GCP
• usable authenticated/API-like paths: Runpod, Vast.ai
• public-page ingestion candidates: CoreWeave, Lambda
• weaker first-wave automation candidates: OCI and quote-driven vendors

That means the project works well as a live GPU price aggregator, but it should not pretend every major provider exposes equally clean machine-readable price feeds from day one.

Current implementation status:

• live now: Azure retail prices API
• next straightforward additions: GCP Billing Catalog API, Runpod API, Vast.ai API
• later/manual-ingestion tier: AWS spot/on-demand blending, CoreWeave, Lambda, OCI

See docs/OPEN_SOURCE_MODEL.md and GOVERNANCE.md.

Quick start

cd /Users/sebastianboehler/Documents/GitHub/compute_atlas
nvm install 22
nvm use 22
corepack enable
cp .env.example .env
docker compose -f infra/docker/docker-compose.yml up -d db
pnpm install
pnpm db:migrate
pnpm db:seed
pnpm dev

To ingest the first live provider and populate real observations:

ENABLE_PUBLICATIONS=false pnpm worker:pipeline

For a containerized web-only preview that matches the production service shape:

docker compose -f infra/docker/docker-compose.yml up --build web

For the full containerized stack with Postgres and worker:

docker compose -f infra/docker/docker-compose.yml --profile full up --build

Useful commands

pnpm lint
pnpm typecheck
pnpm test
pnpm build
pnpm db:migrate
pnpm db:seed
pnpm solana:keypair:create
pnpm publish:devnet
pnpm verify:devnet
pnpm verify:e2e

Environment variables

• DATABASE_URL: Postgres connection string.
• TIMESCALE_ENABLED: Enables Timescale-specific migration steps when the extension exists.
• ENABLE_PUBLICATIONS: Enables downstream publication jobs when explicitly needed.
• SOLANA_RPC_URL: Solana devnet RPC endpoint.
• SOLANA_RPC_WS_URL: Solana websocket endpoint.
• SOLANA_KEYPAIR_PATH: Relative path from the repo root for the publisher keypair.
• SOLANA_AIRDROP_MIN_BALANCE_SOL: Auto-airdrop threshold used by verification.
• SOLANA_AIRDROP_TARGET_BALANCE_SOL: Desired balance after airdrop.
• GCP_BILLING_API_KEY: Optional API key for the next provider tier.
• RUNPOD_API_KEY: Optional API key for later Runpod ingestion.
• VAST_API_KEY: Optional API key for later Vast.ai ingestion.
• ENABLE_ADMIN_UI: Enables the admin page.
• NEXT_PUBLIC_REPOSITORY_URL: Repository link shown in the UI.

Local services

• Web/API: http://localhost:3000
• Postgres: localhost:55432

Cloud Run split

• compute-atlas-web: public web and API service
• compute-atlas-worker: request-driven pipeline job

Deploy assets live in infra/cloudrun.

Community files

• CONTRIBUTING.md
• CODE_OF_CONDUCT.md
• SECURITY.md
• GOVERNANCE.md
• docs/ROADMAP.md

Current limitations

• Azure live retail pricing is implemented first; the other providers are still staged behind future adapters.
• Some large providers expose public APIs, while others will need authenticated APIs or pricing-page ingestion before they can join the board.
• In-memory rate limiting and caching are suitable for single-instance deployments, not scaled multi-instance production.