Loading...
Loading...
Data Lagoon
Define columns once, set a cadence, and agents run discover, validate, and merge in waves. Your live table grows row by row, versioned and verified.
[how it works]
One setup step, then the same loop on the cadence you choose.
Name your table, set columns, pick a cadence. Seed from Hermosa's verified catalog, upload a file, or start empty and let agents discover.
table name
LA Specialty Roasters
entity label
Roaster
slug
la-specialty-roasters
seed from · optional
or start empty · upload CSV · point agents at URLs
schema
cadence
Set once. Waves run on schedule until you change it.
Agents search for new rows, cross-referencing against what is already live. Sources are scraped, parsed, and matched to your table schema.
>wave 8 · table: la-specialty-roasters
>query: "la specialty coffee roasters"
>14 candidates found
>dedup vs baseline (44 rows): 3 new
>
An independent pass fact-checks every candidate against primary sources. Provenance attached, conflicts flagged, confidence scored.
✓stereoscope <- stereoscopecoffee.com
✓roaster · Long Beach, CA
✓3 primary sources agree
>verdict: pass (confidence 0.94)
>
Verified rows land in the live layer, versioned and queryable. Counts update, history recorded, ready to export or build on.
✓3 rows -> live (atomic)
✓snapshot v47 committed
✓reef: +5 edges (roasts, located-in)
>export: CSV · Parquet ready
>
Steps 2 through 4 repeat on a cadence you set: daily, weekly, or manual. Each run is a wave. The next wave picks up where the last left off.
[layers]
Every row passes through three layers before it ships. Candidates sit in raw, fact-checking moves them to clean, verified rows land in live.
[raw] candidates
Stereoscope Coffee
[pending]Roaster · Long Beach, CA
src: stereoscopecoffee.com
[clean] validated
Stereoscope Coffee
[pass · 0.94]Roaster · Long Beach, CA
3 primary sources ✓ · hours ✓
[live] production
Stereoscope Coffee
[v42 · live]Roaster · Long Beach, CA
queryable ✓ · provenance ✓
✕ rejected · blue-bottle-echo · duplicate of Blue Bottle
[raw] Unverified candidates waiting for review, each logged with its source and timestamp.
[clean] Fact-checked against primary sources, ready to merge with full provenance.
[live] Versioned, queryable source of truth. Every change is a snapshot you can roll back to.
[multi-tenant]
Seed from our catalog, upload a CSV, or start empty. Your lagoon, your schema, your cadence.
Pull verified catalog rows in as a seed baseline.
Run waves to dedupe, validate, and extend what you imported.
Upload CSV or point agents at URLs on your cadence.
verified tables · import to process
12 tables · 400+ verified rows
LA Specialty Roasters
[processing]seeded from Third Wave Coffee · 38 rows imported
wave 1 · dedupe · validate · extend with LA-only rows
your schema · your cadence · export anytime
[use]
Structured data you own. Browse it, export it, or build on it.
| Name | City | Website | Hours | Status |
|---|---|---|---|---|
| Blue Bottle | Oakland, CA | bluebottlecoffee.com | 7a–6p | [verified] |
| Intelligentsia | Chicago, IL | intelligentsia.com | 7a–7p | [verified] |
| Stereoscope | Long Beach, CA | stereoscopecoffee.com | 8a–4p | [new in v47] |
use it anywhere · export snapshots, query with SQL, or build custom views and apps on your live table.
Export
CSV, Parquet, or any snapshot
Query
SQL and API on live + history
Build
Grids, dashboards, internal tools
[harness]
Waves run from Cursor, Claude Code, or any MCP harness. You bring inference. We hold storage, versioning, and the graph.
{
"mcpServers": {
"hermosa": {
"type": "streamable-http",
"url": "https://hermosaai.com/mcp",
"headers": {
"Authorization": "Bearer hermosa_key_…"
}
}
}
}read
›list-lagoon-topics
›get-topic-schema
›get-dedup-baseline
›query-lagoon-entities
›get-reef-neighbors
›search-lagoon
write
›submit-proposals
›submit-validation
Connect
Point your agent at the Lagoon MCP server. Tools load automatically.
Run locally
Discovery and validation use your subscription, on your machine.
Merge remotely
Verified proposals land as atomic, versioned snapshots.
[time travel]
Query any version, export that point in time, or diff what changed between waves.
+8 rows (manual seed)
+19 rows (first wave)
+9 reef edges
+13 rows · 22 reef edges
~2 updated (hours fields)
+3 rows (stereoscope + 2)
~1 updated (hours field)
·cadence: weekly
[reef]
Relationships surface during discovery. Follow a roaster to the beans behind the cup.
Under the hood
Define a table once. Run waves on repeat. Export anywhere or build interfaces on your live data.