Bricksearch — Databricks

Ask a question.
Get a research doc.

Multi-agent research across internal intel, customer signal, competitive landscape, and quantitative data — verified by DBRA and curated Genie Spaces — synthesized into a shareable Google Doc.

One-time setup — run these three commands in terminal:

First-time plugin users

git clone https://github.com/databricks-eng/plugin-marketplace.git ~/plugin-marketplace

Copy

Install plugin

isaac plugin add db-bricksearch@experimental

Copy

Create /bricksearch command

cp ~/.claude/plugins/cache/experimental-plugin-marketplace/db-bricksearch/*/commands/bricksearch.md ~/.claude/commands/

Copy

Already installed? Run isaac plugin update to get the latest version

Then type /bricksearch in any Isaac session.

For best results, enable Google and Glean MCPs — they power doc publishing and internal search. Run isaac configure mcp to select servers, or visit the MCP connections page in your workspace. See go/mcp for setup docs.

Step 1

Question triggers multi-agent searches

Bricksearch breaks your question into targeted sub-queries, confirms the plan with you, then dispatches specialized agents in parallel. Each covers a distinct source domain — from internal Logfood data and customer notes to competitor docs and academic standards. DBRA runs alongside for independent quantitative verification. Every agent follows a strict source protocol: nothing is silently skipped.

Agent	Covers	Known bias
DBRA	Autonomous Logfood SQL via Databricks Research Agent — runs twice (dispatch + verification)	May hallucinate table names
Quantitative	Curated Genie Spaces, metric views, DBRA, direct SQL — adoption, spend, trends, funnels	Instrumented features only
Internal Intel	PRDs, UXR archive, PM customer notes, ES tickets, CUJs, prior design work	Past internal thinking
Sales & GTM	BrickBites, SFDC win/loss, QBR themes, PMM messaging, field playbooks	Deal-winning narratives
Roadmap	Quarterly pre-reads, Aha!, Jira epics, leadership priorities, OKRs	Formally planned work only
Competitive	Battlecards, competitor docs, feature matrices, pricing, UX comparisons	Databricks-favorable framing
Market Landscape	Gartner, Forrester, IDC, MAD landscape (~930 companies), VC funding	Analyst views lag 6–18 months
Community Voice	Databricks forums, Reddit, Stack Overflow, Ideas Portal, GitHub issues	Vocal power users
Official Docs	Databricks docs, release notes, blog posts, API references	Shipped features only
Product Design	NNGroup, design systems, competitor UX patterns, accessibility, IA	UX elegance over feasibility
Foundations	Academic papers, W3C/ISO/NIST standards, canonical definitions	Theoretical

Findings are verified, conflicts resolved

As agents report back, findings are cross-referenced and ranked by authoritativeness. DBRA re-queries Logfood to verify any quantitative claim from a non-authoritative source. Contradictions are resolved — not hidden. Gaps trigger targeted follow-up that you approve before re-dispatch.

When sources conflict, the higher-ranked source wins:

Genie Space metric views + Logfood data

Curated metrics & observed behavior

Internal PRDs & strategy docs

Intent & decisions

Official Databricks docs

Publicly committed

Customer notes / Win-loss data

Direct customer signal

Industry standards / Field signal

Established definitions

Community & Ideas Portal

Public signal

Competitive battlecards

Useful, not neutral

External analysts

Independent but lagging

External blogs

Opinion-based

Ask a question.
Get a research doc.

Question triggers multi-agent searches

Findings are verified, conflicts resolved

A shareable Google Doc

Ask a question.Get a research doc.

Question triggers multi-agent searches

Findings are verified, conflicts resolved

A shareable Google Doc

Ask a question.
Get a research doc.