Numbers without definitions
An agent writing SQL against a raw warehouse returns plausible numbers, and nobody can tell which ones are wrong.
Redroot answers from certified metrics, or refuses.
Enterprise AI platform · Built in Jakarta
AI that works on what your enterprise actually knows.
LapisAI gives your agents three governed layers: Redroot for the numbers in your warehouses, Redwood for the knowledge in your documents, and Relai for the work your people do with AI. All of it runs in your environment.
lapis/ˈla.pɪs/Indonesian for layer.
LapisAI platform3 layers
RELAIWorkspace, Code and Desktop. Where people and agents do the work.
Redwood
Documents become linked, cited company knowledge.
Redroot
Certified metrics from every warehouse, asked for by name.
Your cloud · Indonesian region · On-premises
01/04One question, answered through every layer.
Certified metricsCited knowledgeGoverned agentsRuns in your cloudIndonesian regionMCP for any AI toolBahasa Indonesia firstOpen formats, no lock-in
Why LapisAI
Your AI tools can write. They still cannot see your business.
Documents without connections
Search finds passages, not who did what, with whom, and what was learned. Every question starts from zero.
Redwood compiles them into linked, cited knowledge.
Tools without governance
Five AI subscriptions per person, no audit trail, and company data sitting outside your security perimeter.
Relai puts every model and agent in one governed place.
The platform
Three layers. One governed context for every agent.
Each layer stands on its own. Together, an agent moves from a client's revenue to its contracts to a finished brief in one conversation, with a receipt for every fact.
Lapis 03 · AI platform
RELAI
RELAI
Where people and agents do the work: a governed AI workspace, a coding agent and a desktop agent, all grounded in the layers below.
relai.lapisai.id →
Lapis 02 · Ontology
Redwood
What the company knows: documents compiled into linked, cited notes on people, companies, projects and domains, growing with every project.
redwood.lapisai.id →
Lapis 01 · Data platform
Redroot
What the numbers are: a certified semantic model over every warehouse, so agents ask for metrics by name and get the same answer every time.
redroot.lapisai.id →
Foundation · Deployment
Your environment
Every layer deploys where your data already lives: your cloud tenant, an Indonesian region, or your own data centre.
Deployment →
Redroot
Lapis 01 · Data platform
Early access
Certified numbers from every warehouse.
Redroot reads your warehouse's data model and drafts a semantic model your data owner certifies. Agents then ask for metrics by name. Redroot writes the SQL, applies the asker's access policies, and returns the number with its receipt.
- AI drafts, people certify. Every metric carries a named owner and a certification stamp.
- Refuses instead of guessing. Questions outside the certified model get the closest certified metric, not a plausible wrong number.
- The warehouses you already run. SQL Server, Oracle, PostgreSQL, Microsoft Fabric and Synapse first; Snowflake, BigQuery, Databricks and MaxCompute next.
Why a semantic layer: in dbt Labs' April 2026 benchmark, questions answered through a modelled semantic layer were 98 to 100% correct, while raw text-to-SQL on the unmodelled schema reached 64.5%.
Answer · via RedrootSample data
Berapa NPL per cabang Q3 dibanding Q2?
2,41%▲ 0,18 pp vs Q2
NPL gross, all branches, Q3 2026
CertifiedNPL gross · certified by Risk · data as of 08.00
ClaudeRedroot connected
hanya Jawa Barat
NPL gross, Jawa Barat, Q3 2026
2,73% ▲ 0,28 pp
Tersertifikasi
Redwood
Lapis 02 · Ontology
In production at Xquisite AI
A company brain that grows a ring every year.
Redwood turns your documents into an open wiki of linked people, companies, projects and domains, with a source cited for every fact. Agents browse it, search it and draft inside it under each user's permissions, and every project adds to what the company knows.
- Compiled, not just indexed. PDFs, Word files, slide decks and scans become entity notes with links and citations.
- Agents that work in context. Each project gets an agent that drafts deliverables from its plan, meetings and files.
- Yours to keep. The whole vault is plain Markdown you can open in Obsidian or export at any time.
1,608linked notes
1,928graph links
368companies mapped
Redwood's Second Brain, built from Xquisite AI's own pipeline and delivery files.
Entity note · RedwoodSample note
Company
Reviewed
PT Contoh Manufaktur
Manufacturing client since 2024.1 Data warehouse modernisation delivered with Microsoft in 2025.2 Lessons from the handover now guide two newer projects.3
Related
Sources
1Proposal_2024-03.pdf · page 2
2Handover_2025-11.pptx · slide 3
3Lessons learned · DWH Modernisation 2025
![A cited answer in Indonesian on a phone, with source [2] open in a bottom sheet.](assets/redwood/phone.webp)
RELAI
Lapis 03 · AI platform
AI you can rely on.
Where your people and agents do the work.
Relai brings the models, tools and agents your teams need into one governed platform, connected to Redwood and Redroot, so every answer starts from what your company knows.
Relai Workspace
AvailableOne AI workspace for the whole company: chat across models, deep research with citations, slides, image and video, and an organisation knowledge base. SSO, roles and audit built in.
Chat · Research · Slides · Image · Video · Knowledge
relai.lapisai.id →Relai Code
NewAn agent that reads, writes and runs code in your repositories, from the terminal or your editor, with your team's context and guardrails.
$ relai "add NPL by branch to the risk report"
Reading 14 files · plan ready for review
>
relai.lapisai.id/code →
Reading 14 files · plan ready for review
>
Relai Desktop
NewThe same agent in a desktop app for Windows and macOS. Start a task, approve each edit, and review every change before it lands.
Read src/components/SignupForm.tsx
Edited 1 file · +9 −0
Waiting for your approval
Deploys on Alibaba Cloud, Microsoft Azure or BytePlus, with the models each one offers.
How the layers work together
One question, three layers, one answer you can check.
A director asks Relai
“Kenapa NPL Jawa Barat naik di Q3, dan apa yang sudah dibahas dengan tim regional?”
Why did NPL in West Java rise in Q3, and what has already been discussed with the regional team?
1Redroot
Returns certified NPL by branch for Q2 and Q3, filtered to the regions the director may see.
2Redwood
Finds the regional review minutes, the accounts discussed and who owns them, each with a citation.
3RELAI
RELAIDrafts the brief for the board pack, with every number and claim linked back to its source.
Governance and deployment
Built for regulated Indonesian enterprise.
Runs where your data lives
Your cloud tenant, an Indonesian region, or your own data centre. Nothing leaves without your say.
Permissions travel with the data
Every answer respects the asker's access in the source system, and never hints at what they cannot see.
A receipt for every answer
Certified definitions, citations and data freshness on every number and claim, with a full audit log behind them.
Open to any AI tool
Redroot and Redwood speak MCP, so Claude, Microsoft Copilot, ChatGPT and your own agents can use them too.
Bahasa Indonesia first
Every surface works in Indonesian and English, with local number, date and currency formats.
Yours to keep
Open formats throughout: plain Markdown for knowledge, standard SQL for data. No lock-in.
Deploy on
Alibaba CloudMicrosoft AzureAWSBytePlusOn-premises
Company
From the team that builds Indonesia's data warehouses.
LapisAI (PT Lapis Kecerdasan Buatan) is a member of Xquisite AI, the Jakarta data and AI consultancy behind data platforms for more than 90 enterprises.
90+enterprise clients served by Xquisite AI
5sectors: banking, BUMN, manufacturing, logistics, FMCG
See the platform on your own data.
A working session with our team, on your warehouse, your documents and the question your board keeps asking.