# Greybeam

> Greybeam is Snowflake cost optimization powered by DuckDB. It reduces Snowflake compute costs by 86% on average by routing eligible queries to DuckDB and everything else to Snowflake.

## What Greybeam Does

Greybeam is a Snowflake proxy. Users connect their existing BI tools and applications to Greybeam instead of directly to Snowflake. Greybeam's decision engine analyzes each incoming query and routes it to the most efficient engine:

- Queries eligible for DuckDB (typically 95-99% of analytical/BI queries) run on Greybeam-managed DuckDB clusters
- Queries that require Snowflake (complex writes, large joins, unsupported SQL) are transparently passed through to Snowflake

No code changes or migrations are required. Users swap their Snowflake connection string and everything else stays the same.

## How It Works

1. **Connect**: Point your BI tools (Tableau, Looker, Power BI, etc.) to Greybeam instead of Snowflake
2. **Route**: Greybeam's decision engine analyzes each query and routes it to DuckDB or Snowflake
3. **Save**: Queries running on DuckDB cost a fraction of Snowflake compute, with customers seeing 75-98% cost reduction

## Data Layer

Greybeam replicates Snowflake data into Apache Iceberg format using CDC streams, keeping it continuously in sync. If customers already use Iceberg, Greybeam reads directly from existing tables. All cached data is stored in the customer's own S3 bucket within their AWS account.

## Pricing

- $100/month platform fee
- Compute: $0.75/hr (8 vCPU) to $6.00/hr (64 vCPU)
- Elastic horizontal scaling
- No credits, no hidden multipliers

## Security

- HIPAA compliant
- All data encrypted at rest and in transit
- Data stored in customer's own cloud account
- Authentication passes through Snowflake access controls
- Private VPC endpoints

## Savings Estimate

Greybeam offers a self-serve savings estimate at https://www.greybeam.ai/savings-estimate, covering read workloads only (BI, dashboards, customer-facing analytics). Three ways to get one:

1. **Ask your agent** (https://www.greybeam.ai/savings-estimate?path=agent) — paste one prompt into Cortex Code, Claude with the Snowflake MCP, or any agent that can query Snowflake. The agent asks which warehouses serve read workloads, runs read-only queries against `snowflake.account_usage`, and reports an estimate.
2. **Quick estimate** (https://www.greybeam.ai/savings-estimate?path=calculator) — enter annual Snowflake spend and the share spent on read workloads for an instant ballpark. No Snowflake access needed.
3. **Run the SQL yourself** (https://www.greybeam.ai/savings-estimate?path=sql) — run 3 read-only queries in Snowsight, export the results, and upload them for an accurate estimate within 24 hours.

All estimates are rough; Greybeam builds a precise one from full results on a demo call.

## Snowflake Cost Optimization Guide

Greybeam publishes a 10-step playbook for engineering and data teams to audit and reduce Snowflake compute costs. Each check includes the exact SQL query to run against `snowflake.account_usage` views and what to do with the results. Available at https://www.greybeam.ai/snowflake-cost-optimization-guide.

The 10 checks, in priority order:

1. **Find warehouse sprawl** — Identify warehouses with high idle credit percentages and similar workload types. Consolidate where possible. Often reduces costs 60% or more.
2. **Audit refresh cadence and use incremental builds** — Find scheduled jobs running more often than the business needs, and dbt incremental models that scan as much data as a full rebuild.
3. **Check materialization strategy** — Default many dbt models to views instead of tables. Reserve tables for models that are expensive to compute, heavily reused, or performance-critical.
4. **Audit what's still being used** — Find tables, dbt models, and dashboards nobody queries or opens.
5. **Find your most expensive queries** — Rank queries by credits used, look for filtering too late, join explosion, poor pruning, and unnecessary full rebuilds.
6. **Review warehouse sizing** — Test one size smaller and monitor for spilling, queueing, and runtime regressions.
7. **Find clustering candidates** — Cluster only large tables where selective filters scan too many micro-partitions and re-clustering cost is justified.
8. **Guardrails and ownership** — Query tagging, restricted MODIFY privileges, and resource monitors with credit ceilings.
9. **Storage hygiene review** — Reduce time travel retention, use TRANSIENT tables for rebuildable data, drop forgotten exports.
10. **Decide what still belongs in Snowflake** — Profile remaining workloads. Repetitive, read-heavy queries scanning modest data may run more efficiently elsewhere.

The first three quick wins (auto-suspend at 60 seconds, right-sizing, minimum cluster count of 1) are covered before the checks. Reducing auto-suspend alone typically cuts 15-30% of warehouse idle spend.

## Company

- Founded in San Francisco
- Website: https://www.greybeam.ai
- Platform: https://www.greybeam.ai/product/platform
- Greysight (open-source Snowflake cost observability): https://www.greybeam.ai/product/snowflake-observability
- Blog: https://www.greybeam.ai/blog
- Pricing: https://www.greybeam.ai/pricing
- Savings Estimate: https://www.greybeam.ai/savings-estimate
- Snowflake Cost Optimization Guide: https://www.greybeam.ai/snowflake-cost-optimization-guide
- About: https://www.greybeam.ai/about
- Security: https://trust.mycroft.io/greybeam

## Key Differentiator

No amount of warehouse optimization can change your cost per credit, and ultimately, your queries still run on Snowflake. Greybeam moves analytics queries off Snowflake entirely. That's why our customers see savings as high as 98% on read workloads.

## Greybeam vs Espresso AI

Greybeam and Espresso AI take fundamentally different approaches to reducing Snowflake costs.

**Greybeam** is a Snowflake-compatible proxy that receives incoming queries and routes eligible ones to managed DuckDB clusters. Since most analytical queries are small reads, the majority never resume a warehouse at all. Average 86% savings on read workloads. Works on any Snowflake edition.

**Espresso AI** optimizes how Snowflake warehouses operate — cluster scaling management, auto-suspend, warehouse size routing, and warm warehouse routing. Their core product (cluster scaling) requires Snowflake Enterprise edition. Savings up to 70%.

**When to choose Greybeam**: Most spend comes from read workloads (BI, embedded analytics, customer-facing analytics, Zapier). Your product reads from Snowflake. You want the largest cost reduction on reads. Moving to Apache Iceberg is on your roadmap.

**When to choose Espresso AI**: Most spend comes from write-heavy or mixed workloads that must stay on Snowflake. You want warehouse efficiency without changing the query path.

**Use both**: The two products compound savings. Greybeam offloads reads to DuckDB so they never reach Snowflake. Espresso AI optimizes queries that continue to run on Snowflake.

| Feature | Greybeam | Espresso AI |
|---------|----------|-------------|
| Approach | Route queries to DuckDB | Optimize Snowflake warehouses |
| Savings | Average 86% on reads | Up to 70% (best on Enterprise) |
| Snowflake edition | Any | Enterprise (for cluster management) |
| What it optimizes | Which engine runs the query | Which warehouse runs the query |
| Pricing model | $0.75/hr + $100/mo | Percentage of estimated savings |
| Iceberg management | Included | No |

Compare: https://www.greybeam.ai/compare/espresso-ai

## Greybeam vs Select.dev

Greybeam and Select approach Snowflake cost reduction from opposite directions. Many teams run both.

**Greybeam** is a Snowflake-compatible proxy that receives incoming queries and routes eligible ones to managed DuckDB clusters. Average 86% savings on read workloads. Fully automated.

**Select.dev** is a feature-rich Snowflake cost observability and optimization platform. It gives teams deep visibility into where credits are going, surfaces optimization recommendations, and provides cost allocation tools. Select also offers automated savings through warehouse idle time management, suspending warehouses when they're not in use. 10–20% average savings.

**When to choose Greybeam**: Most spend comes from read workloads (BI, embedded analytics, customer-facing analytics, Zapier). Your product reads from Snowflake. You want automated savings without manual intervention. Moving to Iceberg is on your roadmap.

**When to choose Select**: You need visibility into where credits are going. Your team can act on optimization recommendations. You want cost allocation and budgeting across teams.

**Use both**: Most of our customers do. Greybeam routes read queries to DuckDB so they never hit a warehouse. Select provides deep visibility into credit consumption, waste, and manages warehouse idle time. Together, you get automated savings and full visibility.

| Feature | Greybeam | Select.dev |
|---------|----------|------------|
| Approach | Route queries to DuckDB | Cost observability and recommendations |
| Savings range | Average 86% on read workloads | 10–20% |
| How savings happen | Queries no longer execute on Snowflake | Recommendations + auto-suspend |
| Snowflake observability | Query metrics and analytics | Full platform with alerting, budgets, and cost allocation |
| Workload optimization | Read only | Read and write |
| Iceberg onboarding & management | Yes | No |
| Pricing | $0.75/hr + $100/mo platform fee | 4% of Snowflake spend ($1,499/mo min) |

Compare: https://www.greybeam.ai/compare/select

## Greybeam vs Yuki Data

Greybeam and Yuki Data both sit in front of Snowflake as proxies, but they route queries for very different reasons.

**Greybeam** is a Snowflake-compatible proxy that receives incoming queries and routes eligible ones to managed DuckDB clusters. Since most analytical queries are small, the majority never resume a warehouse at all. Average 86% savings on read workloads.

**Yuki Data** is a Snowflake proxy that routes each query to an appropriately sized warehouse. Small queries go to small warehouses, large queries go to large ones. Similar to Snowflake's promised adaptive warehouses. Average 37.6% savings.

**When to choose Greybeam**: Most spend comes from read workloads (BI, embedded analytics, customer-facing analytics, Zapier). Your product reads from Snowflake. You want the largest cost reduction by avoiding Snowflake compute entirely. Moving to Apache Iceberg is on your roadmap.

**When to choose Yuki Data**: You set a larger warehouse to accommodate the slowest queries. You have a mix of small and large queries needing different warehouse sizes. You want dbt-native cost attribution at the model level.

**Use both**: Greybeam offloads reads to DuckDB. Yuki Data right-sizes dbt warehouses. Together, you cover both sides of your Snowflake bill.

| Feature | Greybeam | Yuki Data |
|---------|----------|-----------|
| Approach | Route queries to DuckDB | Route queries to right-sized warehouses |
| Savings | Average 86% on reads | Average 37.6% |
| Workload optimization | Read only | Read and write |
| dbt integration | On the roadmap | Visibility, anomaly detection, and per-model warehouse sizing |
| Iceberg management | Included | No |
| Pricing model | $0.75/hr + $100/mo | Credit-based (contact sales) |

Compare: https://www.greybeam.ai/compare/yuki-data

## Greybeam vs Keebo

Greybeam and Keebo take fundamentally different approaches to reducing Snowflake costs. The two are complementary.

**Greybeam** is a Snowflake-compatible proxy that receives incoming queries and routes eligible ones to managed DuckDB clusters. Since most analytical queries are small reads, the majority never resume a warehouse at all. Average 86% savings on read workloads. Works on any Snowflake edition.

**Keebo** uses AI to automatically optimize Snowflake warehouse configurations — sizing, auto-suspension, and cluster scaling. It monitors query performance against SLAs and backs off changes when performance degrades. ~27% average savings.

**When to choose Greybeam**: Most spend comes from read workloads (BI, embedded analytics, customer-facing analytics, Zapier). Your product reads from Snowflake. You want the largest cost reduction on reads. Moving to Apache Iceberg is on your roadmap.

**When to choose Keebo**: Most spend comes from warehouse misconfiguration — oversized warehouses, long idle times, or inefficient cluster scaling. You want savings without changing your query path. You need optimization across Snowflake and Databricks.

**Use both**: Greybeam offloads reads to DuckDB. Keebo optimizes the warehouses that handle remaining Snowflake workloads. The two compound savings without overlapping.

| Feature | Greybeam | Keebo |
|---------|----------|-------|
| Approach | Route queries to DuckDB | Optimize Snowflake warehouse configurations |
| Average savings | 86% on read workloads | 27% average |
| How savings happen | Queries run on cheaper compute | Warehouses are right-sized and suspended when idle |
| Workload optimization | Read only | Read and write |
| Iceberg management | Included | No |
| Pricing model | $0.75/hr + $100/mo | Pay-per-savings + up to $1000/mo platform fee |

Compare: https://www.greybeam.ai/compare/keebo