Company profile · 2 min read
Embrasure: a warehouse and memory layer for AI agents
Embrasure combines DuckDB queries, Apache Iceberg storage, realtime ingestion and memory for agents working over company data.
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
Published · Updated
Embrasure gives agents a warehouse, catalog and memory surface for answering questions against business data.
What it does
The company homepage shows sources, tables, sync status, queries and a workspace memory surface. Pricing lists Free, Starter at $49/month, Growth at $249/month and Scale at $749/month. The quickstart begins with connecting one source and exploring its catalog.
Buyer and task
Teams that need agents to query business data while keeping source catalogs, lineage and permissions visible.
Workflow boundaries
The checked public pages do not verify connector reliability, query correctness or security controls in a live workspace. Pricing and supported workflows are source claims.
What I would ask
Ask which systems can be connected, how permissions and lineage travel into an agent answer, and what a customer can export if it leaves.
Why it fits
Embrasure is built around the data layer an agent needs before it can answer business questions reliably. The homepage shows connected sources, tables, sync status, metrics and a memory layer, then names realtime ingestion, Apache Iceberg storage, query execution, MCP and CLI access. The quickstart says to connect one source with read-only scope, sync its catalog, verify the physical table identity and explore it before granting wider access. The sources page lists 29 connection options, including Postgres, Stripe, Snowflake, Salesforce, HubSpot and product-analytics systems.
That is a much more concrete starting point than “give your agent your data.” A team can pick one source, verify lineage and test whether a known metric answer points to the expected table and definition. Pricing provides a staged path: Free, Starter at $49/month, Growth at $249/month and Scale at $749/month, with source, storage, scan and AI-credit allowances shown on the page. The allowances are marked as preview and may change.
The boundary remains important. The checked public pages do not verify connector reliability, query correctness or security in a live workspace. A buyer should test permissions, stale data, a schema change and an export path before using the memory layer for consequential decisions.
Quick facts
| Field | Sourced detail |
|---|---|
| Buyer | Teams building agents over business data |
| Inputs | Warehouses, databases, SaaS tools and pipeline sources |
| Agent paths | MCP, CLI and API over cataloged, lineage-aware data |
| Plans | Free; Starter $49/mo; Growth $249/mo; Scale $749/mo, with preview allowances |
Sources checked
official Speedrun profile · company homepage · source page · source page · source page
Sources checked — September 20, 2026.
About the author
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
I cofound Lazyweb and publish Mudpie. This is an owner-written publication, not an independent testing organization. Research notes distinguish observations, sourced reporting and editorial judgment.
First1000 ↗ · X ↗