Turn Millions of Rows Into Governed Truth & Hidden Gold.
Project Keystone is a lightweight, customizable, AI-first datalake. Ask questions about your business in natural speech with the built-in LLM, connect any chosen model via MCP (OpenAI, Claude, Gemini, or locked-down in-house LLMs), and get the AI to build dashboards, reports, and automations—all backed by RBAC, a live Data Map, and full audit logs.
High-Velocity & Enterprise Sources
- Polymarket & Kalshi Feeds
- News & Social Streams
- Stock Exchange & Market Data
- Enterprise DBs CRM & ERP
- SaaS APIs & Spreadsheets
PROJECT KEYSTONE
Governed ingestion, semantic models, built-in LLM & MCP server
- Ingestion
- Data Map
- RBAC
- Audit Logs
- AI & MCP Core
AI Actions, Automations & MCP Connections
- Natural Speech Built-In AI Q&A
- AI Dashboards & Reports
- AI Automations & Workflows
- MCP Server Any Chosen LLM
- Hidden Gems & Data Map
Millions of Rows
Battle-tested daily across prediction markets, news, social & financial feeds.
Any LLM via MCP
Connect OpenAI, Anthropic Claude, Google Gemini, or locked-down in-house LLMs.
Built-In Natural AI
Ask questions in natural speech and let AI build dashboards, reports & automations.
Run Anywhere
Deploy in your own environment (on-prem), your cloud account, or hybrid.
- Cyber Essentials Plus certified
- Aligned with ISO 27001
- Built to comply with POPIA and GDPR
- Deploy On-Premise, Cloud, or Hybrid
Why We Built Project Keystone
Why we created Project Keystone
With our internal projects like AI Predictions and NewsGPT, we were ingesting millions of rows of data every day from Polymarket, Kalshi, global news wires, social media streams, and stock exchange data.
We needed a way to manage that velocity, connect conflicting schemas, and extract the hidden gold inside the data so our AI agents and teams could act on real signals instead of noise.
Traditional enterprise data stacks were bloated, rigid, and bolted AI on as an afterthought. We needed something lightweight, customizable, and AI-First from day one.
So we built Project Keystone: complete with RBAC, full audit logs, a high-velocity ingestion layer, a live data map, a built-in LLM for natural speech Q&A, universal MCP connectivity to any LLM, and AI that builds custom dashboards, reports, and automations.
Where Keystone Cut Its Teeth
Millions of Rows / DayHigh-velocity multi-source ingestion under real market pressure
- PolymarketPrediction Markets
Real-time order books, contract probabilities, volume spikes, and market resolution streams.
- KalshiRegulated Event Markets
High-frequency event contract ticks, macroeconomic market shifts, and liquidity depth.
- Global News FeedsUnstructured & Structured
Thousands of breaking news wires, regulatory filings, and multi-source articles every hour.
- Social Media StreamsSentiment & Velocity
High-noise social signals, narrative velocity, and emerging market-moving conversations.
- Stock Exchange DataFinancial Tick & OHLCV
Equities, indices, macro indicators, and cross-asset correlations alongside event markets.
AI Predictions
A no-code platform for building autonomous AI agents that trade prediction markets. Ingests millions of rows from Polymarket, Kalshi, news, social media, and stock exchanges to surface trading edges in real time.
Live at aipredictions.aiNewsGPT
An AI-generated news channel where autonomous reporting agents—powered by custom-built, fine-tuned LLMs and Project Keystone—verify facts, synthesize multi-source feeds, and produce live coverage.
Live at newsgpt.aiFar More Than Reporting · True AI-First Architecture
Connect any LLM via MCP, talk to your data in natural speech, and let AI build for you.
Project Keystone isn’t a passive data warehouse. It comes with a built-in LLM interface and a governed Model Context Protocol (MCP) server—so your team can ask questions in plain speech, connect the AI models you already use, and have AI build dashboards, reports, and automations directly.
Connect Any Chosen LLM via MCP
Plug Project Keystone directly into the models your organization already uses or trusts via our built-in Model Context Protocol (MCP) server—with RBAC and audit logging enforced on every call.
- OpenAI (ChatGPT / GPT-4o / o-series)
- Anthropic Claude (Claude Desktop, Claude Code & Enterprise API)
- Google Gemini (Gemini Pro / Flash & Vertex AI)
- In-house, locked-down, self-hosted or air-gapped local LLMs
Ask Questions in Natural Speech
Use Keystone’s built-in AI assistant to talk directly to your business data in plain English—no SQL, no waiting days for an analyst queue.
- Ask complex cross-department questions in everyday natural language
- Uncover hidden gems, correlations, outliers, and margin shifts across millions of rows
- Every answer is calculated from governed semantic metrics—never hallucinated
- Click any number in an AI answer to trace its exact lineage back to raw source rows
AI That Builds Dashboards, Reports & Automations
AI in Project Keystone doesn’t just chat—it builds and acts. Tell the AI what you need monitored or automated, and it constructs the views, reports, and workflows for you.
- Build live interactive dashboards and custom charts from a simple voice or text prompt
- Generate recurring board-ready reports and executive briefings automatically
- Create AI-driven automations, threshold triggers, and multi-system workflows
- Every automation and agent action runs inside your permission boundaries with a full audit log
Model-Agnostic via Governed MCP Server
Bring your preferred cloud frontier model or run 100% locked-down in-house LLMs:
Key Features · Everything You Need, Zero Bloat
Engineered to manage your data, govern access, and find the hidden gems.
Project Keystone replaces fragmented pipelines, unverified spreadsheets, and risky AI wrappers with six tightly integrated capabilities.
Built-In AI, Universal MCP & Automations
Connect any LLM via MCP, ask questions in natural speech, and let AI build dashboards, reports, and automations.
Project Keystone is built for AI from the ground up. Use the built-in LLM to query your business in natural speech, or connect OpenAI, Anthropic Claude, Google Gemini, or your own locked-down in-house LLMs via MCP. Go beyond Q&A by having the AI build live dashboards, custom reports, and automated workflows.
- Universal MCP Server: Connect OpenAI, Anthropic Claude, Google Gemini, or locked-down local LLMs safely.
- Natural Speech Q&A: Use the built-in LLM to ask questions about your business and uncover hidden gems in plain language.
- AI Builder: Get the AI to build interactive dashboards, custom reports, and end-to-end business automations.
- Zero-Hallucination Guardrail: The semantic engine computes the exact math; the AI never invents figures or touches unguarded raw tables.
High-Velocity Ingestion Layer
Ingest millions of rows from any source—with every single row accounted for.
Connect databases, prediction markets, financial feeds, SaaS APIs, webhooks, event streams, and manual spreadsheets. Every extraction runs with declared watermarks and lookback windows so late-arriving data is never missed.
- Strict Run Accounting invariant: rows_read = inserted + skipped + excepted—nothing is silently dropped.
- Immutable, append-only raw landing zone so models can be rebuilt from raw evidence at any time.
- Sandboxed file & stream parsers with strict memory/size bounds and zero store credentials.
- Automated freshness monitoring and an exception ledger that flags schema drift or late feeds immediately.
Interactive Data Map & Flow Lineage
Understand every data flow and trace any figure back to its raw source record.
Most companies have no idea where a dashboard number actually came from. Project Keystone gives both executives and engineers a live, interactive Data Map showing every source, transformation, business entity, and downstream consumer.
- End-to-end visual lineage from raw source tables through Bronze, Silver, and Gold models to final dashboards and automations.
- Identity resolution spine that connects the same customer, asset, or event across disconnected systems.
- Shared semantic layer: define a business metric (like Revenue, Margin, or Active Customers) once, and every dashboard, automation, and AI agent uses that exact definition.
- Click any number to inspect its definition version, freshness timestamp, and underlying records.
Granular RBAC & Runtime Confinement
Access is enforced at the moment a question is asked—for humans and LLMs alike.
Security is the boundary of the semantic layer, not an afterthought bolted onto a BI tool. Whether a user opens a dashboard, runs an automation, or queries via Claude/OpenAI over MCP, Project Keystone enforces the exact same Role-Based Access Control (RBAC).
- Fine-grained Role & Permission model checked per request—with row-level and column-level security.
- Automatic PII classification & masking: sensitive fields stay masked unless the principal holds explicit reveal permission.
- Population threshold guards: refuses queries on tiny cohorts that could expose individuals by subtraction.
- Two-person rule (separation of duties) enforced in transactions for irreversible actions like data erasure.
Full Immutable Audit Logs
Every query, MCP call, dashboard build, permission grant, and AI automation is permanently recorded.
When regulators, internal audit, or the Board ask who accessed sensitive records or why an AI automation triggered an action, Project Keystone gives you an immediate, tamper-evident answer.
- Complete audit trail of sign-ins, governed reads, MCP tool calls, PII reveals, role changes, and automated actions.
- Machine credential attribution that detects if an API key is used from multiple locations simultaneously.
- Dedicated, separately encrypted subject-reference mapping for clean POPIA / GDPR right-to-erasure compliance.
- Built by a Cyber Essentials Plus certified engineering team aligned with ISO 27001.
Lightweight & Deployable Anywhere
Run it in your environment, in your cloud account, or hybrid.
You shouldn’t have to hand your company’s crown-jewel data over to a multi-tenant SaaS black box. Project Keystone is lightweight, modular, and engineered to run inside a boundary you control.
- On-Premise / Your Environment: Deploy on your own servers, private data center, or single-server Docker stack—paired with locked-down in-house LLMs.
- Your Cloud Account: Stand up on AWS, GCP, or Azure inside your own organization policy and KMS encryption keys.
- Hybrid Topology: Keep sensitive operational and PII stores on-premise while bursting analytical or AI workloads to the cloud.
- Full handover option: We can operate it for you as your partner, or hand over the keys with zero remaining vendor credentials.
Interactive Example · One Governed Truth
Switch between an Enterprise Boardroom scenario and our High-Velocity AI Market scenario:
Scenario A · Enterprise Executive Truth
“How many paying customers do we actually have?”
Contacts, accounts, support profiles, and billed subscriptions are all different counts. Instead of debating four conflicting spreadsheets in Exco, Project Keystone resolves identities across systems and applies one agreed definition everywhere.
What Project Keystone enforces here:
- • Agreed business entity & metric definition applied once across all consumers
- • Explicit ordering & deduplication rules—never guessing state from arrival time
- • Click-through lineage from the summary figure back to every contributing source row
Executive Diagnostic
Do you actually need a datalake?
Not every company needs a datalake. If all your data lives in a single app and your reports already agree, keep it simple. But if these six problems sound familiar, there is a specific place for Project Keystone in your business.
Sales, Finance, and Operations walk into Exco with different numbers.
Two board packs both call a figure "Revenue" or "Active Customers," but use different cutoff dates, refund rules, or account statuses—so the meeting starts with an argument over whose spreadsheet is right.
With Project Keystone
Define the metric once in Keystone’s Semantic Layer. Every dashboard, report, automation, and connected LLM reads through that single agreed calculation.
Result: Spend the Exco meeting discussing what changed in the business, not debating the spreadsheet.
You are sitting on millions of rows of data, but missing the hidden gold.
High-frequency market feeds, customer transactions, IoT telemetry, news, and social signals arrive every minute, but human analysts only have time to look at top-level averages.
With Project Keystone
Keystone ingests high-velocity streams continuously and lets you use built-in AI or your chosen LLM via MCP to spot hidden correlations, regime shifts, and outliers—and build automations to act on them.
Result: Spot margin leaks, market shifts, and revenue opportunities before your competitors see them.
You want to connect OpenAI, Claude, Gemini, or an in-house LLM to your data safely.
Connecting an LLM directly to raw production databases risks leaking sensitive PII, breaking access rules, and generating hallucinated numbers that nobody can verify.
With Project Keystone
Connect any chosen LLM via Keystone’s governed MCP server—or use the built-in LLM in natural speech. Every call runs under strict RBAC, masks PII automatically, and logs every action.
Result: Give your team and leadership natural-speech AI and automated workflows that your CISO and Auditors can sign off on.
Building new dashboards, reports, and automations takes weeks of engineering time.
Every time leadership asks a new question or wants an automated alert/workflow, someone has to file a ticket, write custom SQL, and wire up another script.
With Project Keystone
Ask Keystone’s AI in natural speech to answer the question, build the dashboard, schedule the custom report, or set up the automation directly on top of governed metrics.
Result: Go from a business question in natural speech to a live dashboard or automation in minutes.
The customer or operational story is split across five disconnected silos.
CRM knows the leads, Billing knows the invoices, Support knows the tickets, and Operations knows the delivery—but answering one cross-business question takes days.
With Project Keystone
Keystone’s Identity Resolution Spine links records across systems with scored confidence, building one unified timeline across every department.
Result: Ask questions about the entire customer or asset lifecycle in seconds.
A dashboard number drops, and nobody can explain why.
A KPI moves sharply and leadership cannot tell whether the business actually changed, an upstream API feed is three hours late, or records were rejected during import.
With Project Keystone
Source freshness, run accounting (read = inserted + skipped + excepted), and full row lineage sit inside the same platform as your dashboards and AI tools.
Result: Know immediately whether you are looking at a real business event or a late upstream feed.
You DO need Project Keystone if…
- Your business data sits across 3+ systems (ERP, CRM, Billing, Support, Spreadsheets, or external market feeds) that don’t agree.
- You want to connect OpenAI, Claude, Gemini, or an in-house LLM via MCP, ask questions in natural speech, and let AI build dashboards, reports, and automations safely.
- You are ingesting high-volume records and suspect there are hidden margin leaks, anomalies, or revenue opportunities buried inside.
- You require strict RBAC, full audit trails, POPIA/GDPR compliance, and the ability to run in your own environment, cloud, or hybrid.
You probably DON’T need a datalake yet if…
- All of your operations run inside a single SaaS tool whose built-in dashboard already answers every question you have.
- You have no need to combine records across departments or trace historical changes over time.
- Your problem is actually a missing operational workflow app rather than a data and AI problem (in which case, our Custom Software team at Vexalabs will tell you honestly and build that instead).
Nothing Hidden · Full Data Map & Flow
See exactly what happens when you ask a question or trigger an automation.
Project Keystone is not a black box. Every layer—from read-only source extraction and run accounting to the semantic security boundary, MCP server, and AI builder—is mapped and auditable.
Click any layer in the stack to see how data flows upward from your systems and how a natural-speech question or MCP call travels through RBAC, semantic definitions, and source lineage.
01 · You ask a question in natural speech, trigger an AI automation, or connect your LLM via MCP.
Talk to your data in natural speech using the built-in LLM, instruct AI to build live dashboards, custom reports, and automations, or inspect the interactive Data Map and Operator Console (job health, freshness, exception ledger, access admin).
Sovereign Deployment · On-Premise, Cloud, or Hybrid
Run it in your environment, your cloud, or hybrid.
A deployment stands inside an environment you control—under your organization policy, your encryption keys, and your audit boundary. We engineer the governance, the connectors, the semantic models, and the AI & MCP integrations.
- Deploy on-premise (private servers / Docker), in your cloud (AWS, GCP, Azure), or hybrid
- Connect OpenAI, Anthropic Claude, Google Gemini, or locked-down in-house LLMs via MCP
- Your sources read safely, with every single row accounted for
- RBAC and PII masking enforced on every dashboard, natural-speech query, and automation
- Operated with Vexalabs as your partner, or handed over with zero vendor lock-in
- Container Runtimes
- Analytical Store (Bronze/Silver/Gold)
- Operational Control Store
- Immutable Raw Object Storage
- Secrets Manager
- Identity & RBAC Engine
- Audit & Exception Ledger
- Ingress, Egress & WAF
- Built-In LLM & MCP Server
- Customer-Managed KMS Keys
- AI Dashboards & Automations
- Automated Retention & Erasure
Every store, secret, cryptographic key, audit log, and AI pipeline lives inside a security boundary you control.
On-Premise / Private Infrastructure
Maximum data sovereignty inside your own walls—with in-house locked-down LLMs.
Deploy Project Keystone directly onto your own Linux servers, VMware/Proxmox cluster, or air-gapped private data center. Pair it with your own locked-down local LLMs so not a single byte of data or AI prompt ever leaves your building.
Key Capabilities:
- Zero data leaves your private network boundary
- Connects to in-house, locked-down local LLMs via built-in UI or MCP
- Single-server Docker profile or multi-node orchestrated cluster
- Full hardware and cryptographic key control
Dedicated Cloud Deployment (AWS / GCP / Azure)
Stood up in your cloud account, not a shared vendor SaaS.
We provision Project Keystone inside an AWS, Google Cloud, or Azure account that your organization owns—using your KMS encryption keys, your VPC/networking rules, and your chosen cloud or enterprise LLMs (OpenAI, Claude, Gemini).
Key Capabilities:
- Provisioned via automated infrastructure-as-code in your account
- Encrypted at rest with your own customer-managed KMS keys
- Multi-zone high availability for queries, dashboards, and automations
- Optional full handover: revoke our deployment role when ready
Hybrid & Vexalabs-Managed
Keep sensitive stores local while scaling compute & AI in the cloud.
Keep raw PII, subject-reference maps, or legacy database extractors inside your private environment while running analytical models, market/news ingestion, or cloud LLMs in a dedicated enclave—operated by your team, Vexalabs, or together.
Key Capabilities:
- Strict egress policy & encrypted service-to-service tunnels
- PII pseudonymisation before data crosses environment boundaries
- Co-managed with Vexalabs Outsourced CTO & Data Engineering support
- Seamless migration path between on-prem and cloud profiles
Boardroom & Exco Readiness
The questions your Exco & Board will ask—answered up front.
We sit with Exco and the Board as an Outsourced CTO and engineering partner. We know the exact questions your CEO, CFO, CIO, and Risk Officer will ask before approving a data and AI platform.
“Why build on Project Keystone instead of buying Snowflake, Databricks, Fivetran, dbt, and a separate BI & AI stack?”
Stitching together five enterprise data vendors costs hundreds of thousands in annual licenses and requires a dedicated data engineering team just to maintain the glue—before you get a single answer. Project Keystone is lightweight, unified, and AI-first from day one: ingestion, run accounting, data map, semantic layer, RBAC, audit logs, natural-speech AI, MCP server, and AI-built dashboards and automations all live in one coherent platform.
“Which LLMs can we use, and can we run locked-down in-house models?”
You have complete freedom. Project Keystone includes a built-in LLM interface so users can ask questions in natural speech and instruct the AI to build dashboards, reports, and automations out of the box. In addition, Keystone’s built-in MCP (Model Context Protocol) server lets you connect OpenAI, Anthropic Claude, Google Gemini, or your own in-house, locked-down local LLMs—all governed by the exact same RBAC and audit rules.
“How do you guarantee AI won’t hallucinate numbers or run rogue automations?”
Whether you use the built-in AI or connect Claude, OpenAI, or Gemini via MCP, the LLM is never allowed to run arbitrary unguarded SQL or invent math. All AI interactions execute through Keystone’s governed semantic layer under the user’s RBAC permissions. Every figure in an AI-generated dashboard, report, or narrative is verified against the semantic layer, and every automation runs inside explicit permission and approval boundaries.
“How do we prevent AI and cloud costs from spiralling out of control as our data volume grows?”
Keystone is engineered around strict cost controls. Deterministic engines compute the heavy aggregations on governed models, while the LLM handles natural-speech intent, dashboard/automation specs, and concise synthesis—so AI token costs never scale with millions of raw rows. Furthermore, every table declares an explicit retention policy so storage never grows unbounded.
“How does Keystone handle POPIA, GDPR, PII, and insider threats?”
Every store operation is confined to a separate runtime identity. PII columns are classified in the model and masked by default unless a user holds explicit reveal permission—meaning sensitive PII never enters an external LLM’s context window. Personal identifiers are pseudonymised via a separately encrypted subject-reference map, and irreversible operator actions require two-person approval.
“Who owns the platform, the data, and the IP—and are we locked into Vexalabs?”
Your data, your business metric definitions, and your deployment live inside your own environment or cloud account under your own encryption keys. All domain models and metric definitions are stored as clean, human-readable declarations in your repository. Vexalabs can operate and evolve the platform for you as your technical partner, or hand over the keys completely.
How We Roll It Out · Fast, Practical & Proven
Three steps from scattered data to AI-ready intelligence.
We don’t start with an 18-month science project. We deploy quickly, connect your key data sources, and prove the value on your real business data.
- 01
Connect your priority sources
We identify your core operational databases, APIs, spreadsheets, or market feeds and connect them to Project Keystone with read-only extractors.
- 02
Stand up Keystone in your boundary
We deploy Project Keystone in your environment, cloud account, or hybrid setup, map your data flows, and configure your core metrics, RBAC rules, and audit logs.
- 03
Connect your LLMs, dashboards & automations
Ask questions in natural speech with the built-in LLM, connect OpenAI, Claude, Gemini, or in-house models via MCP, and let the AI build dashboards, reports, and automations.
Ready to get started?
Put Project Keystone to work on your data.
Talk to our technical team about deploying Project Keystone in your environment, cloud, or hybrid—and connecting your business data to natural-speech AI, MCP, dashboards, and automations.