The graph as the primary surface
Engines, missions, workflows, integrations, and data sources are nodes. Selecting a node opens the Inspector, which is how you navigate the system rather than a side panel you occasionally consult.
React Flow
BLKPVNTHR.OS Live
(opens os.blkpvnthr.com in a new tab)
BLKPVNTHR.OS
“The graph is the application — every node is clickable, opens the Inspector.” That is the product’s own in-app tagline, and it is the literal design of the system.
BLKPVNTHR.OS is decision intelligence infrastructure: it connects fragmented data, evidence, analytical engines, workflows and human review into systems designed to support traceable, data-backed decisions. Applications, workflows, integrations, experiments, data sources, and system capabilities are modeled as connected resources on an interactive knowledge graph — so a conclusion can be linked back toward the evidence and the steps that produced it.
Live at os.blkpvnthr.com — opens in guest mode, no sign-up.
at a glance
The system in one screen
- Role
- Sole designer, engineer, and operator.
- Status
- Live and public. Opens in guest mode.
- Interface
- React 19, TypeScript, Vite, Tailwind v4, React Flow, Zustand, React Router 7.
- Services
- Managed backend (auth + 6 edge functions), FastAPI research engine (Python).
- Integrations
- Read-only bank links, paper-only brokerage, managed backend.
- Infrastructure
- A reverse proxy fronts the system; the research engine binds to loopback and is not exposed directly.
- Quality
- Vitest, oxlint.
the problem
Context is lost at every boundary
Data sources, research systems, automation, knowledge, and operational workflows are typically fragmented across disconnected applications. The data lives in one place, the analysis in another, the notes in a third, and the decision nowhere at all.
Each boundary between those applications is a place where context disappears. A hypothesis loses its link to the data that produced it. A metric loses its link to the decision it was meant to inform. A workflow loses its link to the engine it was meant to trigger. The work of holding the whole picture together falls back on the person, every single time.
BLKPVNTHR.OS is a personal answer to that fragmentation, built and operated by one engineer. It is not investment advice, and it is paper-first.
the approach
A connected decision-intelligence model
Rather than another dashboard, BLKPVNTHR.OS models everything it knows about as a resource, and every relationship between those resources as an edge. The graph is not a visualization bolted onto the app. The graph is the app.
Extensible across data-intensive domains
Domain-specific engines connect specialized data, models, tools, and workflows to the shared decision infrastructure. New systems can be introduced with their own routes and operational boundaries without rebuilding the platform around each use case.
/engines/*
Terminal-style interaction
The landing surface is a terminal at root@blkpvnthr-os:~$. Typing start
opens the AI Brain. Commands are a first-class way into the system, not a hidden power-user mode.
Command interface
Evidence and provenance
Evidence and provenance are modelled explicitly. The architecture is designed so conclusions can be linked to their supporting data, experiments, assumptions, and processing steps, and so outputs distinguish sourced values, calculations, assumptions, and generated interpretation.
evidence_fusion
Modular workflows
Workflows, missions, and an event bus compose engines into repeatable operations, so automation is assembled from parts that already exist rather than written from scratch each time.
/workflows
Personal and organizational modes
The same operating model is designed to serve an individual tracking a single domain, and can extend to an organization running many — laboratories, schedules, and pipelines — on one connected graph.
Guest mode available
capabilities
What the system actually does
BLKPVNTHR.OS gives a decision system the parts it needs: connecting fragmented data and context, evaluating evidence and assumptions, running analytical and scenario-based workflows, preserving provenance and decision history, comparing possible actions, and producing reviewable outputs for the people who make the call. Each domain can define its own data sources, analytical engines, evidence requirements, workflows, and review boundaries on the shared graph-based architecture.
current implementation
The current live deployment uses financial analysis, quantitative research, and paper-trading workflows as a working implementation of the broader decision-intelligence architecture. These modules demonstrate how the platform connects source data, analytical processes, evidence, review boundaries, and recorded actions within one navigable system.
Financial analysis
Connected account data supports reporting, goal tracking, and reviewable financial insights through read-only workflows.
/engines/finance
Research and validation
Hypotheses, experiments, evidence, and validation stages are modeled as connected resources so findings remain tied to the work that produced them.
/research-division
Controlled action workflows
Research, risk review, and execution remain separated, demonstrating how the architecture can place explicit controls between analysis and consequential action.
/engines/trading
The same architecture can support additional decision systems in areas such as operations, workforce planning, technical research, program management, risk analysis, and other data-intensive environments. Each implementation can define its own data sources, analytical engines, evidence standards, workflows, permissions, and human-review requirements without rebuilding the underlying platform.
platform capabilities
Knowledge and operations
Interactive knowledge graph
The AI Brain at /os. Every node is clickable and opens the Inspector.
Memory and context
A memory module and a knowledge base hold context across sessions and across engines.
Workflow orchestration
Workflows, missions, an event bus, a Meta-Agent, and a research scheduler coordinate the engines.
Command interface and navigation
Terminal, Graphify, Vault, Email, Career, and Automation are all reachable as part of one system.
/graphify · /vault · /memory
architecture
Three layers
The interface talks to orchestration, and orchestration composes the engines. Credentials stay server-side — nothing in the browser holds a provider secret.
The following diagram describes three layers. Each layer is built on the layer below it: the interface layer is built on the orchestration layer, and the orchestration layer is built on the engines layer.
-
Interface
- Terminal root@blkpvnthr-os:~$
- AI Brain (graph) React Flow
- Application views & Inspector React 19 · Zustand
-
Orchestration
- Workflows /workflows
- Event Bus
- Meta-Agent
- Missions /missions
- Memory /memory
-
Engines
- Live data sources & ingestion
- Research /research-division
- Knowledge Knowledge Base
- Action controlled execution
- Automation
stack
Technical stack
Interface
- React 19
- TypeScript
- Vite
- Tailwind v4
- React Flow (@xyflow/react)
- Zustand
- React Router 7
- Framer Motion
- Recharts
- Radix UI
Services
- Authentication and access control
- Serverless edge functions × 6
- FastAPI research engine
- Python (stdlib-first)
- research_engine
- pattern_engine
- alpha_discovery
- evidence_fusion
- strategy_lab
Data & integrations
- Read-only financial data integration
- Brokerage connectivity (demo)
- Managed application database
Infrastructure
- Reverse proxy
- Loopback-bound research engine
Quality
- Vitest
- oxlint
boundaries
Trust and safety boundaries
Read-only financial data by default
You sign in with your bank inside Plaid’s secure widget. The app never sees your bank credentials, and the Plaid secret stays server-side. Balances only: no transfers.
Paper-first trading
Trading is paper-first. Live orders stay disabled until a strategy clears the paper-trading threshold, and the Alpaca proxy targets the paper API.
Analysis is separate from execution
Analysis and execution permissions are kept separate, so a research module cannot reach the execution path on its own.
Source-backed analysis
The system is designed not to fabricate financial values, and to trace figures back to the account data or the research run they came from.
Human review for consequential decisions
Anything with a real consequence is surfaced for review rather than acted on silently.
User-controlled credentials
Connections are established, inspected, and removed by the user. Nothing is linked in guest mode.
These are product design boundaries, not a formal compliance certification. Nothing in BLKPVNTHR.OS is investment advice.
the system
Inside BLKPVNTHR.OS
The current live implementation includes finance-oriented engines and workflows. These screens demonstrate how domain-specific data, analytical tools, evidence, and actions can be organized through the shared decision-intelligence architecture.
Six screens from the current live implementation. Scroll, swipe, or use the arrows.
-
AI Brain. Every engine, mission, workflow and data source is a node; selecting one opens the Inspector. -
AI CFO Department. An auditable accounting engine that closes the books each month and tracks long-term net worth. Figures shown are the product’s clearly-labeled demo data, not real finances. -
Quantitative Research Division. 18 laboratories and a research lifecycle that runs from observation through statistical validation to a deployable strategy. Paper and research only. -
Trading engine. Paper-first: live orders stay disabled until a strategy clears the paper-trading threshold. Any figures shown are the product’s own demo state — not real finances, and not a track record. -
Memory System. Seven memory types — user, project, research, conversation, semantic, working and RAG — that the agents create, tag, link and search across runs. -
Loop Engineering. Every run feeds the next — scheduled routines log their execution, compound lessons, and close the loop. Simulated runtime; the persistence and UI are real.
start assembles the AI Brain.
Any values visible are the product’s own demo state, not real finances.
maturity
Where it stands today
BLKPVNTHR.OS is an actively developed software platform and working product environment. Some capabilities are production-oriented, while others remain experimental research modules.
The fastest way to judge it is to open it. It runs in guest mode with no sign-up.
disclosure
BLKPVNTHR.OS is a separate service
BLKPVNTHR.OS is a separate software service operated as a distinct product. It may be governed by its own product-specific terms and privacy disclosures, which are separate from the Terms of Use and Privacy Policy that apply to asmaa.dev. Review those before using it.
Nothing in the product is investment advice. Trading is paper-first and live orders are disabled. Bank connections are read-only through Plaid — balances only, no transfers. See the full Financial Disclaimer for details.
Explore the live system
Launch BLKPVNTHR.OS to explore the interface, architecture, and connected operating model.