rebyteai Agent Context — Data for Agents
Agent Context is a managed data layer inside your Agent Computer. It gives you:
- Organization data — databases, S3 files, and services your team has connected
- Your own tables — import files as queryable datasets scoped to your workspace
- SQL queries — standard SQL (Apache DataFusion) across all sources
- Views — materialized joins across datasets, auto-refreshed on source changes
Setup
curl -fsSL https://cc-tools-binaries.s3.amazonaws.com/rebyte-data/rebyte-data-linux-amd64 -o /usr/local/bin/rebyte-data && chmod +x /usr/local/bin/rebyte-data
Auth is automatic — credentials are pre-configured in this VM.
Quick Start
# See what data is available
rebyte-data tables
# Query a table
rebyte-data query 'SELECT * FROM "customers" LIMIT 10'
# Import your own file
rebyte-data import ./results.csv --name results
# Query your imported data
rebyte-data query 'SELECT category, COUNT(*) FROM results GROUP BY category'
Commands
| Command | Description |
|---|---|
rebyte-data tables |
List all datasets you can access |
rebyte-data schema <table> |
Show column names and types |
rebyte-data query "SQL" |
Execute a SQL query, print results as table |
rebyte-data import <file> [--name <table>] |
Upload a file and create a queryable dataset |
rebyte-data drop <table> |
Delete a dataset you created |
Importing Data
Import CSV, Parquet, JSON, JSONL, or TSV files. Schema is auto-inferred.
# Import with explicit name
rebyte-data import ./sales_2024.csv --name sales
# Name is auto-derived from filename if omitted
rebyte-data import ./quarterly_report.parquet
# → creates table "quarterly_report"
After import, the data is immediately queryable via SQL. Your imported tables are scoped to your workspace — other agents can't see them.
Supported formats: .csv, .tsv, .parquet, .json, .jsonl
SQL Reference
Standard SQL via Apache DataFusion.
Always double-quote table names — identifiers are case-sensitive:
-- ✅ Correct
SELECT * FROM "my_dataset" LIMIT 5
SELECT c.name FROM "customers" c JOIN "orders" o ON c.id = o.customer_id
-- ❌ Wrong
SELECT * FROM my_dataset LIMIT 5
Common patterns:
-- Filter and aggregate
SELECT region, SUM(revenue) FROM "sales" GROUP BY region ORDER BY SUM(revenue) DESC
-- Join across sources (org data + your imported data)
SELECT o.id, o.total, r.category
FROM "orders" o
JOIN results r ON o.product_id = r.id
-- Text search
SELECT * FROM "products" WHERE name ILIKE '%widget%'
-- Date filtering
SELECT * FROM "events" WHERE created_at > '2024-01-01'
Semantic Search
Datasets with embedded columns support vector similarity search:
-- Find similar documents
SELECT title, cosine_similarity(embedding, embed('quarterly revenue trends')) AS score
FROM "documents"
ORDER BY score DESC LIMIT 5
-- Combine structured filters with semantic search
SELECT * FROM "support_tickets"
WHERE status = 'open'
ORDER BY cosine_similarity(embedding, embed('billing issue')) DESC
LIMIT 10
Architecture
- Organization data (admin-managed) — connected databases, S3 buckets, views. Read-only. Always fresh or cached with scheduled refresh.
- Agent data (your imports) — files you upload, stored in cloud storage, queryable via SQL. Scoped to your workspace.
- Views — materialized SQL queries cached in DuckDB. Auto-refresh when source data changes via S3 notifications.
All data is queryable through a single SQL interface. No need to know where the data lives.