DataQubeDataQube

Blog

Notes from building the enterprise agent OS

How we think about agents, custody, provenance, and institutional memory - written by the people building it.

Product28 July 2026·5 min read

Enterprise AI needs institutional memory, not just better models

Enterprise AI is not only about better models. Trusted answers need data access, governance, business context, and institutional memory.

Product Engineering · DataQube

Ask in plain Englishno SQL
CRM
ok
BI
ok
Docs
ok
answer generated · evidence attached
Product25 July 2026

Your company has the answers. Your team just can't reach them.

Most companies have the data they need, but teams still wait on analysts, dashboards, and SQL requests. AI-powered self-service analytics can change that.

Product Engineering · 5 min read

source of truthwarehousedocumentspolicies
AIquestion moves,
data stays
Product25 July 2026

Why AI should come to your data, not the other way around

Every company wants faster answers from AI. The harder question is architectural: should sensitive data move to the AI, or the AI to the data?

Product Engineering · 5 min read

Drill into the Industrials drag…
warehouse.sqlas-of joined✓ evt-8f3a21
/pytop = drilldown(att, level="issuer")
Product21 July 2026

An AI notebook for power users: one kernel, two sets of hands

Most AI analysis tools treat everyone as a business user — answers in, questions out, experts exiled to another tool. DataQube is a shared workbench: the agent and your quants co-work in the same kernel, on the same dataframes, onto the same record.

Product Engineering · 4 min read

Weekly IC PackMon 06:00
Quarterly ESG ScreenFri 07:00
ALCO Stress Pack1st of month
run #38 · parameters pinned✓ matches run #37
Product8 July 2026

Deterministic AI workflows: a language, not a recording

Deterministic AI workflows can't be built by saving a session as a job — repetition needs properties you can read before anything runs. Why DataQube workflows are written in a language: ordered steps, typed parameters, and a grant manifest that can't lie about what it touches.

Product Engineering · 5 min read

Drill into the Industrials drag…
warehouse.sqlas-of joined✓ evt-8f3a21
/pytop = drilldown(att, level="issuer")
Engineering18 June 2026

Context compaction without amnesia

Every long AI investigation outgrows the context window — the honest question is what happens then. How DataQube separates the model's working context from the append-only record, makes compaction a visible event, and keeps compacted threads fully auditable.

Engineering · 5 min read

DataQube Data Gateway14 toolsfirst-party
Market Data9 toolsOAuth
Risk Engine6 toolsworkload id
Product2 June 2026

AI agent permissions: why ours asks first

AI agent permissions are the question under every enterprise deployment: what can it do without a human, and who decided? How DataQube treats capability as a grant, stops mid-run to ask, and puts every denial on the record.

Product Engineering · 6 min read

network posture
outbound routes0
telemetrynone
licensesigned file · validated locally
model endpointin-cluster · your GPUs
Engineering12 May 2026

Air-gapped AI agents: what zero egress actually takes

Air-gapped AI deployment is not a feature flag — it is a constraint you either designed for or didn't. Where the model runs (your cloud with zero data retention, or fully inside), how delivery and licensing work with no route out, and what support means when nobody can SSH in.

Platform Engineering · 4 min read

benchmark T+1 lag · validated
Research22 April 2026

Your AI analyst needs a brain, not a vector store

AI agent memory built on similarity search retrieves what sounds related. Institutional knowledge is structural — what applies where, what supersedes what, who validated it. Why DataQube's memory is a knowledge graph with supersession semantics, and how it knows when it's wrong.

Research · 7 min read

evt-8f3a21query.execute1,842 rows✓
evt-a10441permission.grantesg.api · once✓
evt-77d0e3auth.decisionwall-crossing✗ denied
evt-b81f55agent.runweekly-ic-pack✓
Security30 March 2026

Provenance by construction: AI analysis your auditors can replay

An AI audit trail works only if it is built into how answers are produced, not written up afterwards. How DataQube constructs provenance — queries, citations, workflow runs, refusals — so a reviewer can replay any figure without system access.

Engineering · 7 min read

Deploys in your cluster - nothing leaves

See your data answer questions - without leaving your infrastructure

Thirty minutes with an engineer: live product, deployment options, and your security team's questions answered by someone who wrote the code.

Book a demoRead the security overview