DataDesk — AI for Executives · Nionium Inc.

DATADESK / Mark-1

DataDesk is the AI SaaS platform. Mark-1 is the agent. Together: a persistent, memory-equipped AI agent with real-time access to financial markets, geopolitical intelligence, research databases, and 200+ specialized skill modules — each tenant on a dedicated, fully-isolated Ubuntu VPS. Built to think, remember, and act across sessions.

SaaS Platform
Persistent Memory
Multi-Platform
200+ Skills
Real-Time Data
Autonomous
Adaptive Intelligence
Every Mark Is Different
With time, every Mark becomes uniquely shaped by its user. It memorizes preferences, learns patterns, and creates customized internal skills, concepts, and workflows — all designed to serve you better. No two Marks are alike.
How Mark Learns — Real Examples
Week 1
"Book me a hotel in Rome for June."
Mark asks: Budget? Star rating? Location preference? Dates?
Week 8
"Book me a hotel in Rome for June."
Mark already knows: 5-star, city center, Genius discount, king bed, late checkout. Sends 3 options to your Telegram with comparison table.
Same request. Completely different experience. That's compounding intelligence.
Week 1
"What's happening in the markets today?"
Mark gives a generic summary — top headlines from major indices, broad strokes, no personal relevance.
Week 8
"What's happening in the markets today?"
Mark leads with your NVDA and PLTR short positions, flags an SEC filing on ORCL, adds Georgia macro outlook you track, and skips sectors you've never cared about.
Same question. One gives noise, the other gives signal tailored to your portfolio.
Week 1
"Run due diligence on this SaaS company."
Mark asks: What's the company name? Which aspects — financials, team, IP, market? What format do you want? Any specific red flags to look for?
Week 8
"Run due diligence on this SaaS company."
Mark built a custom DD skill from your previous 6 reviews. Auto-pulls SEC filings, runs churn analysis, checks founder backgrounds, benchmarks against your deal criteria, and delivers a 12-section PDF matching your fund's template — with a Telegram voice summary.
From blank-slate questions to a complete investment workflow it wrote itself.
Week 1
"Prepare my Monday board update."
Mark asks: What metrics? Which KPIs? What time range? Who's the audience? What format — slides, doc, email?
Week 8
"Prepare my Monday board update."
Mark auto-runs at 6AM every Monday — pulls revenue from your Sheets, compares WoW/MoM, generates a 5-slide deck in your brand template, highlights anomalies, and drops it in your inbox before your first coffee. Zero prompts needed.
From asking 5 clarifying questions to delivering it before you even ask.
Week 1
"Fix the failing CI pipeline."
Mark asks: Which repo? What CI system? Can you paste the error? What branch?
Week 8
"Fix the failing CI pipeline."
Mark already knows: your GitHub org, the monorepo structure, that CI runs on GitHub Actions, your pytest + xdist setup. Pulls the failing workflow, reads the logs, identifies the flaky test, opens a PR with the fix, and pings you on Telegram when CI goes green.
From "which repo?" to autonomous diagnosis, fix, and PR — end to end.
Persistent Memory Self-Authored Skills User-Shaped Workflows Compounding Intelligence
00 — DataDesk
The SaaS Platform for Executive AI

DataDesk is an AI SaaS platform by Nionium Inc, positioned as "AI for Executives." A premium flat-rate subscription at $980/month — no usage caps, no token limits, no metering. Each tenant gets a dedicated Ubuntu VPS running their own fully-isolated Mark-1 agent with private memory, private skills, and private data. Available at app.mark-1-datadesk.com.

$980/mo Flat
No usage caps. No token limits. No metered API billing. One flat rate, full access.
Dedicated VPS
Each tenant gets their own Ubuntu VPS. Full isolation — CPU, RAM, storage, filesystem, and memory.
Full Stack
Next.js frontend + FastAPI backend + PostgreSQL + Traefik + Docker. Production-grade from day one.
Per-Tenant Encryption
API keys and credentials encrypted per-tenant. Trading keys never touch shared infrastructure.
Platform Modules

Second Brain

SQLite-backed durable knowledge management — pages, feeds, attachments, knowledge graph. Hybrid FTS5 + semantic-vector search with N-citation synthesis.

Knowledge Graph

Markets Panel

Live financial data via FMP, FRED, and DBnomics. Real-time quotes, fundamentals, valuations, 1.7B+ economic time series. Full forex, crypto, commodities, indices.

FMP · FRED · DBnomics

Deep Research

Multi-source research pipeline. arXiv, OpenAlex, PubMed, SEC EDGAR, Exa, Tavily. Synthesized reports with citations across academic, financial, and geopolitical sources.

Multi-Source Pipeline

Alpaca Trading

Per-tenant encrypted Alpaca brokerage integration. Programmatic equity trading through the agent — orders, positions, portfolio management via natural language.

Per-Tenant Encrypted

Screening

Sanctions, PEP (politically-exposed-person), and AML watchlist screening. Real-time compliance checks against global regulatory databases.

Sanctions · PEP · AML

Intel Map

Live geospatial intelligence — real-time flights (ADSB.lol) and vessels (AISStream.io) on an interactive map. OSINT monitoring with geospatial overlays.

Live Flights & Vessels

Morning Briefing

Automated daily synthesis — aggregates unread mail, calendar, Kanban tasks, news, and overnight developments into a single briefing delivered before you wake.

Daily · 7AM

Obsidian Sync

Bi-directional Obsidian vault synchronization. Second Brain pages sync to your local Obsidian vault — your knowledge graph, accessible offline.

Bi-Directional

Prompt Library

Reusable, versioned prompt templates. Store, categorize, and invoke proven prompts across sessions and workflows. Team-shareable within your tenant.

Versioned

My Feeds

Aggregate YouTube channels, RSS feeds, and Telegram channels into a single Second Brain digest. Auto-ingested, searchable, and summarised.

YouTube · RSS · Telegram

MoA Conductor

Multi-model orchestration (Mixture-of-Agents). 3-reference conductor preset: Sonnet 5 + GLM-5.2 + Kimi K2.7 Code, M3 aggregator. Routes prompts across models for superior synthesis.

3-Model Conductor

Build Mode

Plan → execute workflow. The agent writes a structured plan to disk, then executes it step by step with verification. Complex multi-step builds without losing the thread.

Plan → Execute

Voice Mode

OpenAI Realtime API integration. Bidirectional voice conversations with the agent — full duplex, low latency, natural interruption. Powered by OpenAI's Realtime transport.

OpenAI Realtime

Image Generation

ComfyUI and DALL-E integration for text-to-image generation. Composable workflows, style presets, and palette consistency for article illustrations and infographics.

ComfyUI · DALL-E

Admin Dashboard

Tenant administration — user management, billing, project registry (18 projects, 110 versions), system health, cron monitoring, and skill inventory. Full operational control.

Tenant Management

Mark-1 Local

Fully offline coding agent on your laptop. pi-based agent runs against local Ollama models — results stream back through the same DataDesk UI. 100% offline on RTX 5000 Ada.

Offline · Ollama
Platform Architecture
DataDesk Stack
Next.js
Frontend · SSR · React
FastAPI
Backend · Async · Python
PostgreSQL
Per-Tenant Database
Traefik
Reverse Proxy · TLS
Docker
Container Orchestration
↓ ↓ ↓ ↓ ↓
Per-Tenant Isolation
Ubuntu VPS
Encrypted Keys
Private Memory
Private Skills
Private Second Brain
01 — Your Own Mark
Not a Shared Service. A Dedicated Agent — Yours Alone.

Every user gets their own Mark — not a shared instance, not a multi-tenant platform, not a chatbot behind an API. A fully isolated AI agent running on its own Ubuntu VPS with its own file system, its own memory, its own skills, and its own intelligence that compounds exclusively from your interactions. Your data never touches another user's environment. Ever.

Dedicated VPS
Your own Ubuntu server. Not shared infrastructure — a real machine with dedicated CPU, RAM, and storage.
Private Memory
Persistent memory trained only on your data. Your preferences, your corrections, your patterns — no one else's.
Custom Skills
Skills that Mark writes itself — from your workflows, your tools, your domain. They evolve with you, not a user base.
Full Isolation
No shared databases, no cross-contamination. Your agent's data is physically separated from every other Mark in existence.
Architecture — Every Mark Is a Sovereign System
Mark — User A
Ubuntu 24.04 VPS
Private file system
Persistent memory
Custom skill library
Encrypted credentials
Your Mark Active
Ubuntu 24.04 VPS
Private file system
Persistent memory
Custom skill library
Encrypted credentials
Mark — User C
Ubuntu 24.04 VPS
Private file system
Persistent memory
Custom skill library
Encrypted credentials
Speed
No queue. No rate limits from other users. No "the system is at capacity" messages. Your Mark uses 100% of its server's resources — all the time. Long-running tasks, parallel sub-agents, background cron jobs — nothing competes for attention.
Security
Your API keys, credentials, business documents, and strategic conversations live on your server only. No shared database a breach could expose. No other user's prompt injection can reach your data. Physical isolation — not just logical.
Reliability
When another user's agent crashes or overloads, yours keeps running. Independent uptime, independent restarts, independent deployments. Your Mark's 24/7 cron jobs don't stop because someone else did something heavy.
02 — Overview
Intelligence at a Glance
202
Specialized Skills
20
Skill Categories
28
Active Cron Jobs
50+
Tool Integrations
∞
Session Memory
457
Second Brain Pages
1,603
Knowledge Entities
1.7B+
Economic Time Series
03 — Core Capabilities
What Mark-1 Can Do

Financial Markets

Real-time quotes, fundamentals, valuations, insider trading, SEC filings, screeners via FMP. Full forex, crypto, commodities, indices coverage.

FMP API

Economic Data

1.7B+ time series from FRED + DBnomics — interest rates, GDP, CPI, unemployment, yields, housing, money supply, VIX. World Bank and DBnomics for global data.

FRED + World Bank

Deep Research

Academic papers via arXiv, OpenAlex, PubMed. Chemical compounds via PubChem. Protein data via UniProt. Prediction markets via Polymarket.

Multi-Source

News & OSINT

RSS feed monitoring with auto-refresh, blogwatcher-cli integration. Real-time news aggregation across categories with cron-based scanning.

Real-Time

Web Intelligence

Neural/semantic search via Exa AI (10B+ pages). Web extraction, browser automation with stealth. YouTube transcript extraction via Supadata.

Exa AI

Legal & Regulatory

SEC EDGAR access for US securities filings. Georgian legislation via Matsne. Full-text search across regulatory databases.

SEC + Matsne

HTML Artifacts

Interactive dashboards, visualizations, presentations — fully styled HTML with 54 production design systems. Dark themes, responsive, exportable.

54 Design Systems

Infographics

21 layout types × 21 visual styles. Professional infographics, visual summaries, and data visualizations generated from raw data.

441 Combinations

Video & Animation

Manim for 3Blue1Brown-style math animations. ASCII video pipelines. p5.js interactive generative art. Pixel art with retro palettes.

Multi-Format

Diagrams

Architecture diagrams (dark SVG), Excalidraw hand-drawn style, flowcharts, sequence diagrams, concept maps. All exportable.

SVG + Excalidraw

Documents & Slides

PowerPoint generation, PDF manipulation with nano-pdf, Google Docs/Sheets via Workspace API. Notion page creation and management.

Full Suite

Audio & Music

Text-to-speech synthesis. Spotify playback control. Songwriting assistance with AI music generation prompts. Audio visualization.

TTS + Spotify

Full-Stack Development

Write, debug, test, and deploy code. Python, JavaScript, shell scripting. TDD methodology, systematic debugging, code review pipelines.

End-to-End

Multi-Agent Orchestration

Spawn subagents for parallel workstreams. Delegate to Claude Code, Codex, OpenCode. Orchestrate complex tasks across isolated contexts.

Autonomous

ML/AI Operations

Model training (Axolotl, Unsloth, TRL), inference (vLLM, llama.cpp), evaluation (60+ benchmarks), Hugging Face Hub, Weights & Biases.

Full MLOps

DevOps & Infrastructure

GitHub workflow (PRs, issues, code review), webhook subscriptions, server management, cron job scheduling, process monitoring.

CI/CD

Browser Automation

Stealth browsing with bot-detection bypass. CDP control via browser-harness. Form filling, scraping, screenshot analysis, visual QA.

Headless + CDP

Red Teaming

LLM jailbreaking via G0DM0D3 techniques. Refusal removal from open-weight models (OBLITERATE). Security-focused adversarial testing.

Security

Travel Management

Flight search and booking via Duffel (300+ airlines, real fares). Hotel search on Booking.com with Genius optimization. Weather forecasting.

Duffel API

Communications

Gmail read/send/search. AgentMail autonomous inbox. Telegram messaging. Multi-platform message delivery. Himalaya CLI for IMAP/SMTP.

Multi-Channel

Productivity

Google Calendar, Drive, Sheets, Docs, Contacts. Notion database management. Linear issue tracking. Task scheduling with cron jobs.

Workspace

Smart Home

Philips Hue control via OpenHue — lights, rooms, scenes, color temperatures. Home automation triggers and scheduling.

OpenHue

Location Intelligence

Geocoding, reverse geocoding, distance calculation, route planning. Maps integration for spatial awareness and logistics.

Maps API

Autonomous Scheduling

Cron-based task execution with no user present. Chained jobs with context passing. Auto-delivery to Telegram, Discord, SMS, email.

24/7 Cron
04 — Skill Domains
Proficiency Matrix
Skill Distribution by Domain
Capability Depth — Core Domains
Financial Intelligence
97%
Research & OSINT
95%
Software Engineering
90%
Content Creation
88%
ML/AI Operations
87%
DevOps & Infra
85%
Communications
83%
Travel & Logistics
80%
Smart Home / IoT
72%
05 — Skills Inventory
Complete Skill Catalog
Category Skills Count Key Capabilities
Research arxiv, blogwatcher, dbnomics, deep-research-report, equity-research-report, exa-search, felo-search-integration, fred, gdelt-monitor, llm-wiki, matsne-georgia-law, news-roundup, openalex, perplexity-style-search, polymarket, pubchem, pubmed, research-paper-writing, rss-feed-monitoring, sec-edgar, screening, streamlined-research-formatting, tavily-search, uniprot, wikipedia-image-sourcing, world-bank, … 30 Academic papers, economic data, prediction markets, legal databases, biomedical research, deep research reports, OSINT, screening
Creative architecture-diagram, ascii, baoyu-article-illustrator, baoyu-comic, baoyu-infographic, claude-design, comfyui, creative-ideation, design-md, excalidraw, html-artifact-with-charts, html-dashboard-artifact, humanizer, ideation, manim-layout-guardrails, manim-video, p5js, pixel-art, popular-web-designs, pretext, sketch, songwriting-and-ai-music, touchdesigner-mcp, … 26 HTML artifacts, diagrams, animations, infographics, generative art, music, ComfyUI, design tools, humanizer, creative ideation
Productivity airtable, booking-hotel-search, customgpt, datadesk-app, docx, duffel-travel, eur-import, firecrawl-real-estate, fmp-financial-data, google-workspace, household-budget, linear, maps, nano-pdf, notion, ocr-and-documents, perplexity-style-search, powerpoint, teams-meeting-pipeline, travel-lifestyle-concierge, visual-object-comparison, weather, … 28 Financial data, travel booking, Google suite, Notion, project management, real estate, web search, PDF editing, Airtable, CustomGPT, household budget, travel concierge
Software Dev agent-live-context-injection, agent-tool-gating, chat-web-search-optimization, context7, date-time-context-injection, debugging-hermes-tui-commands, engineering-workflow, firecrawl-backend-integration, hermes-agent-skill-authoring, hermes-custom-slash-command, hermes-multi-model-orchestration, large-artifact-targeted-editing, loops, multi-source-research-pipeline, node-inspect-debugger, oss-integration-screening, plan, post-quantum-cryptography, python-debugpy, react-streaming-markdown-images, realtime-integration, reasoning-model-streaming, second-brain-dev, second-brain-think-synthesis, sqlite-cli-to-multi-tenant-api, streaming-agentic-tool-loop, … 26 Engineering workflow, debugging, planning, code review, multi-agent development, streaming, context injection, API design, voice integration, MoA orchestration, PQC, loops, research pipeline
DevOps api-change-watchdog, api-credential-first-touch, autonomous-executive-intelligence, autonomous-loop-discipline, cloudflare-quick-tunnel, datadesk-file-architecture, deployment-confirmation, docker-ops, fullstack-docker-traefik-deploy, hermes-cli-branding-chrome, hermes-cli-clear-renderer-reset, hermes-custom-provider-config, hermes-gateway-ops, hermes-instance-cloning, hermes-safe-update, hostinger-vps-provisioning, iran-report, morning-briefing-pipeline, projects-per-tenant, projects-site-artifact-deploy, server-side-og-image-preview, tailscale-mesh-admin, telegram-integration, telegram-multi-instance-conflict, webhook-subscriptions, … 27 Deployment, Docker, Traefik, Cloudflare tunnels, Hostinger provisioning, Iran SITREP, Morning Briefing, per-tenant ops, Tailscale, Telegram, project artifacts, API watchdogs, autonomous intel
DataDesk datadesk-admin-tools, datadesk-agent-disconnect-troubleshoot, datadesk-billing-integration, datadesk-canvas-editor, datadesk-chat-architecture, datadesk-chat-indicators, datadesk-chat-mode-extension, datadesk-chat-tools, datadesk-feeds-feature, datadesk-image-editing, datadesk-image-mode, datadesk-screening-integration, datadesk-search-image-pitfalls, datadesk-watchlists, hybrid-fast-path-crud, news-headline-sanitization 17 Admin tools, billing, canvas editor, chat architecture, feeds, image mode, screening, watchlists, CRUD patterns, news sanitization
GitHub codebase-inspection, github-auth, github-code-review, github-issues, github-pr-workflow, github-repo-management 6 Full PR lifecycle, issue tracking, code review, repo management
Autonomous AI Agents claude-code, codex, hermes-agent, kanban-codex-lane, opencode, pi-agent 6 Spawn autonomous sub-agents, multi-agent orchestration, offline coding agents
Email agentmail, himalaya 2 Autonomous inbox, IMAP/SMTP management
Media gif-search, heartmula, songsee, spotify, youtube-content 5 Spotify control, YouTube transcripts, GIF search, audio visualization, AI music generation
Note-Taking obsidian, second-brain, second-brain-storage-surfaces 3 Obsidian vault integration, Second Brain knowledge management, storage surface mapping
Gaming minecraft-modpack-server, pokemon-player 2 Minecraft modpack hosting, Pokemon emulation
Other browser-harness, dogfood, godmode, native-mcp, openhue, webhook-subscriptions, xurl, jupyter-live-kernel 8 Browser CDP, smart home, social media, gaming, webhooks, MCP, red teaming, Jupyter kernels
06 — Architecture
System Design
Input Layer — Platforms
DataDesk Web
Telegram
Discord
CLI
Webhooks
Cron Jobs
↓ ↓ ↓ ↓ ↓
Core Engine
MARK-1 — Hermes Agent Core
Kimi K3 · Persistent Memory · Session History · Skill Router
↓ ↓ ↓ ↓ ↓
Tool Layer
Terminal
Browser
File System
Web Search
Vision
↓ ↓ ↓ ↓ ↓
Integration Layer — APIs & Services
FMP
FRED
Duffel
Google
Notion
GitHub
Exa AI
Fireworks AI
Tavily
↓ ↓ ↓ ↓ ↓
Output Layer — Delivery
DataDesk Artifacts
Telegram
Email
Files
HTML Artifacts
Voice / TTS
Intel Map
Live Flights
07 — Why Mark-1
Beyond ChatGPT and Claude

Anthropic and OpenAI provide exceptional conversational AI. What they don't provide is a persistent, autonomous agent that remembers you, acts on tools, carries a growing skill library, schedules cron work while you sleep, and runs long multi-step loops without watching over them. Mark-1 is engineered for that.

M
Memory
4 layers
Cross-session, across projects
T
Tools
5 native + 21 fast-path
Terminal, browser, files, vision
S
Skills
202 skills
20 categories, loaded on demand
C
Cron Jobs
28 active
Briefings, backups, monitors
L
Loops
Multi-step
Subagents, sustained work
01

Four layers of persistent memory — not just a long context window

Conventional LLM providers forget everything between sessions. A 1M-token context window is irrelevant if the next chat starts empty. Mark-1 carries four independent memory layers that survive reboots, VPS migrations, and months of inactivity.

Agent Memory

Cross-session facts about the user, environment, tool quirks, and lessons learned. Auto-injected into every prompt. ~262k chars currently used across ~245 entries.

User Profile

Dedicated profile of the operator — name, communication style, preferences, family, projects. Hand-curated, never trimmed. ~30k chars.

Second Brain

SQLite-backed durable knowledge — pages, feeds, attachments, knowledge graph. Currently 1,603 entities, 4,362 relations, 3,350 mentions across 457 pages. Hybrid FTS5 + semantic-vector search, with N-citation synthesis.

Project State

Per-project versioning, snapshot registry, files tree, schema introspection. 23 registered projects with cumulative ~155+ versions.

02

Real tools — and 21 fast-path interceptors that skip the LLM entirely

Function-calling exists everywhere. What doesn't exist: a tool layer that knows when to NOT call the model at all. Mark-1 intercepts high-frequency query classes at the backend (YouTube transcripts, weather, sports scores, financial quotes, macro indicators, flights, etc.) and pre-fetches the real data before the agent sees the question. The model just synthesises.

Native tools (always available)
  • terminal Full bash with PTY
  • browser Headless Chromium via CDP
  • filesystem Read, write, patch, search
  • web_search Tavily + Exa fallback
  • vision Gemini + M3 multimodal
  • delegation Up to 3 concurrent subagents
Fast-path interceptors (backend)
  • youtube — oEmbed + Supadata transcript
  • image vision — Gemini Flash identity pre-fetch
  • sports — TheSportsDB premium lookup
  • fifa — local JSON read from cron cache
  • fmp — quotes, fundamentals, valuation, news
  • fred / dbnomics — macro series
  • weather — Open-Meteo (no key)
  • duffel — flight offers, timezone-aware
  • tavily visual — destinations with images
  • screening — sanctions / PEP / AML
  • gdelt — global media monitoring
  • financial-juice widget parse, … + more
03

202 specialised skills — loaded on demand, not stuffed into every prompt

Skills are self-contained markdown documents with steps, commands, and pitfalls for a recurring task. The agent loads ONLY the skills relevant to the current question. 202 skills across 20 categories. Conventional providers ship a single "you can do anything" model with zero task-specific procedure cache.

Category Skills (sample) Count
Research arxiv, dbnomics, fred, gdelt-monitor, matsne-georgia-law, news-roundup, openalex, perplexity-style-search, polymarket, pubmed, sec-edgar, tavily-search, world-bank, felo-search-integration, … 30
Software Dev agent-tool-gating, hermes-agent-skill-authoring, plan, react-streaming-markdown-images, streaming-agentic-tool-loop, subagent-driven-development, systematic-debugging, engineering-workflow, … 26
Creative html-dashboard-artifact, manim-video, excalidraw, baoyu-infographic, comfyui, architecture-diagram, claude-design, creative-ideation, … 26
Productivity datadesk-app, fmp-financial-data, google-workspace, linear, notion, powerpoint, ocr-and-documents, weather, duffel-travel, … 28
DevOps hermes-safe-update, docker-ops, kanban-orchestrator, telegram-integration, tailscale-mesh-admin, deployment-confirmation, projects-per-tenant, … 27
DataDesk datadesk-app, datadesk-chat-mode-extension, datadesk-image-mode, datadesk-watchlists, datadesk-billing-integration, … 17
GitHub github-pr-workflow, github-code-review, github-issues, github-repo-management, github-auth, codebase-inspection 6
Autonomous AI Agents claude-code, codex, hermes-agent, opencode, pi-agent, kanban-codex-lane 6

Each skill is a self-updating markdown document — versioned, reviewed, and extended as the system discovers edge cases.

04

Cron jobs — autonomous work that doesn't wait for a user to ask

Mark-1 schedules 28 active jobs that fire in fresh sessions, run their scripts or prompts, and deliver results directly to Telegram, email, or a pinned DataDesk chat session. No human in the loop.

Daily / weekly work
  • Morning Briefing — 7am Tbilisi, Gmail + Calendar + Kanban + news
  • Daily backups — 4am UTC to GitHub, 30-day retention
  • Iran Theater SITREP — daily synthesis + public blog
  • Peptide protocol reminder — 7am Telegram ping
  • Project auto-snapshots — every 6h for 4 active projects
Real-time monitoring
  • News ticker — refresh every 1m
  • FMP / NBG rates — refresh every 2h
  • Flights & vessels — live WebSocket fan-in
  • Second Brain feed digest — every 5m
  • Heartbeat watchdogs — backend restart on env-var change
05

Loops — agentic work that runs to completion, not until impatience

Conventional chat interfaces time out after a single response. Mark-1 sustains 220-turn loops, spawns concurrent subagents, runs verified regression harnesses after every change, and keeps building until the user's actual goal is met.

Subagent fan-out

Three concurrent leaf subagents, scoped to one sub-problem each. Useful for parallel research, multi-file audits, and "find the bug across three services."

Verify-and-fix loops

Every code edit ships a regression harness (`verify-*` script) that runs against live systems, fixes silently failed probes, then re-runs until green.

Autonomous intelligence cycle

30-min morning sweep + 15-min evening watch. Reads news, VERIDIUM OSINT, and Second Brain delta; surfaces high-signal items to chat. T2 calibration before T1 alerts.

Mark-1 Local — fully offline on your laptop

A pi-based coding agent runs entirely on your hardware against local Ollama models. Results stream back through the same DataDesk UI as cloud replies — toggle "LOCAL" in the chat toolbar to route through, no separate panel. Verified at 100% offline on RTX 5000 Ada with qwen3.6:35b at 24 GB Q4.

08 — Integrations
Connected Services
Service Type Status
Financial Modeling PrepFinancial Data● Active
FRED (Federal Reserve)Economic Data● Active
Duffel TravelFlight Booking● Active
Google WorkspaceProductivity● Active
NotionKnowledge Base● Active
Telegram BotMessaging● Active
SupadataYouTube Data● Active
Context7Framework Docs● Active
Browser UseRemote Browser● Active
AgentMailEmail Inbox● Active
TavilyWeb Search & Extract● Active
FirecrawlWeb Scraping & Crawling● Active
Fireworks AILLM Inference● Active
ADSB.lolLive Flight Tracking● Active
NBG (National Bank of Georgia)Exchange Rates● Active
Second Brain APIKnowledge Management● Active
AISStream.ioLive Vessel Tracking● Active
CustomGPT (VERIDIUM)OSINT Knowledge Base● Active
Integration Coverage by Domain
09 — Cognitive Architecture
Memory & Intelligence Model

Persistent Memory

Durable facts survive across sessions — user preferences, environment details, integration configs, conventions. 15,000 character capacity injected every turn.

Always-On

Session History

Full-text searchable transcripts of every past conversation. Keyword and semantic search across all sessions for cross-session recall.

Searchable

Skill System

202+ procedural skills loaded on demand. Each contains step-by-step workflows, pitfalls, API endpoints, and proven patterns. Self-updating when issues are found.

Self-Healing

Cron Scheduler

Autonomous task execution on schedule — news scanning, data collection, monitoring. Jobs run independently with auto-delivery to configured channels.

Autonomous

Sub-Agent Delegation

Spawn isolated worker agents for parallel tasks. Each gets own terminal, toolset, and context. Orchestrator pattern for complex multi-step workflows.

Parallel

Vision & Analysis

Image analysis, screenshot interpretation, visual QA testing. Can read charts, diagrams, documents, CAPTCHAs, and complex visual layouts.

Multi-Modal
10 — Always Listening, Always Learning
From Conversation to Permanent Memory — Automatically

Mark-1 connects to the Limitless Pendant — a wearable AI device that records your virtual calls, in-person meetings, and real-world conversations. Every word is transcribed, structured, and fed directly into Mark's permanent memory. Nothing gets lost. Every decision, commitment, and insight from every meeting becomes searchable context that Mark-1 can reference weeks or months later.

How It Works
1 Record — The Limitless Pendant captures audio from any conversation: Zoom calls, boardroom meetings, phone calls, hallway discussions.
2 Transcribe — Audio is converted to high-fidelity, speaker-labeled transcripts in real time. Names, topics, and action items are identified automatically.
3 Structure — Transcripts are broken into key decisions, commitments, follow-ups, and context blocks — not just raw text.
4 Store — Structured data flows into Mark-1's permanent memory and session history. It becomes part of the agent's long-term intelligence.
5 Recall — Ask Mark-1 "What did the CFO say about Q3 margins in last Tuesday's call?" and get an exact, sourced answer — instantly.
Investor Call
Full transcript stored. Mark-1 auto-extracts term sheet changes, valuation discussions, and follow-up commitments.
Board Meeting
Every strategic decision captured. Ask Mark-1 to draft minutes, track action items, or compare positions across meetings.
Client Meeting
Requirements, pricing discussions, and verbal agreements — recorded and referenced in future conversations automatically.
Brainstorm Session
Hallway ideas, whiteboard sessions, and informal syncs. Nothing dies in someone's memory — it lives in Mark's.
11 — Parallel Intelligence
One Prompt. Multiple Agents. Simultaneous Results.

When a task has independent parts, Mark-1 doesn't process them one by one. It spawns dedicated sub-agents — each with its own tools, workspace, and focus — that work simultaneously. The results are synthesized into a single coherent output. What takes a sequential AI 15 minutes takes Mark-1 three.

Example — "Compare 5 SaaS companies for acquisition"
Traditional AI — Sequential
Company A — Financials, team, IP~4 min
Company B — waiting…~4 min
Company C — waiting…~4 min
Company D — waiting…~4 min
Company E — waiting…~4 min
Total Time ~20 min
VS
Mark-1 — Parallel Agents
Agent 1 → Company A~4 min
Agent 2 → Company B~4 min
Agent 3 → Company C~4 min
Agent 4 → Company D~4 min
Agent 5 → Company E~4 min
Total Time ~4 min
5 companies analyzed simultaneously — 5× faster, same depth per company
"Research the Iran–Israel situation from all angles"
Agent 1 Scans RSS feeds and live news for latest developments
Agent 2 Pulls VERIDIUM OSINT docs on Iran–Israel military posture
Agent 3 Checks Polymarket prediction odds + financial impact on oil/defense
Three agents return independently. Mark-1 merges into a single structured briefing — news, intelligence, and market impact in one view.
"Prepare quarterly investor materials"
Agent 1 Pulls revenue data from Google Sheets, calculates KPIs
Agent 2 Generates a 10-slide PowerPoint deck with charts
Agent 3 Drafts the investor email with highlights and narrative
Deck, data, and email land together. Reviewed and sent in one conversation — not three separate workflows.
"Audit our GitHub repos for security issues"
Agent 1 Scans repo A — dependency vulnerabilities, secrets in code
Agent 2 Scans repo B — same checks, parallel execution
Agent 3 Scans repo C — CI configs, exposed endpoints, permissions
All three repos audited simultaneously. Consolidated report with severity rankings and fix PRs ready to merge.
12 — Nothing Like It
What Google Built for $10M, You Get as a Dedicated Agent.

Individual pieces of Mark-1 exist elsewhere. But no one has assembled them into a single, dedicated, self-improving agent that runs on your own infrastructure. Here's the landscape — and where Mark-1 sits in it.

Capability Mark-1 ChatGPT / Claude / Gemini Devin / Manus / Lindy
Dedicated SaaS platform ✓ DataDesk — $980/mo flat ✗ No platform ✗ No platform
Dedicated isolated VPS per user ✓ Your own Ubuntu server ✗ Shared cloud ✗ Multi-tenant SaaS
Persistent cross-session memory ✓ Compounds over months Limited — resets per chat Partial — basic recall
Self-authoring skill system ✓ 202+ auto-evolving skills ✗ Static knowledge ✗ Fixed capabilities
Autonomous 24/7 operation ✓ Cron jobs, proactive delivery ✗ Requires user prompt Partial — scheduled tasks
Parallel sub-agent spawning ✓ Simultaneous workers ✗ Single thread Partial — limited
120+ live data sources ✓ FMP, FRED, SEC, DBnomics, Exa… ✗ Stale training data Partial — few integrations
Real-world conversation capture ✓ Limitless Pendant → memory ✗ Not available ✗ Not available
Multi-platform delivery ✓ Telegram, email, voice, webhook ✗ Browser only Partial — Slack/email
Full server access ✓ Terminal, file system, git, Docker Partial — temp sandbox Partial — limited shell
Deep web research (Tavily + Firecrawl) ✓ Real-time crawl + extract ✗ Stale training cutoff Partial — basic browsing
Multi-model inference routing ✓ Fireworks AI, Claude, GPT-OSS ✗ Single model only ✗ Fixed model
Live flight & geospatial tracking ✓ ADSB.lol + Intel Map ✗ Not available ✗ Not available

They're Chatbots. Mark Is Infrastructure.

ChatGPT is a conversation. Mark-1 is a server, a memory system, a skill engine, a cron scheduler, and a multi-agent orchestrator — all running 24/7 on your hardware. The gap isn't incremental. It's architectural.

Dedicated System

Day 1 vs Day 365

Every mainstream AI is identically useful on day one and day one thousand. Mark-1 on day 365 knows your portfolio, your travel style, your reporting format, your team, and your strategic priorities. It's a different agent entirely.

Compounding Value

Your Data. Your Server. Period.

When a multi-tenant AI platform gets breached, everyone's data leaks. Mark-1 is physically isolated — your credentials, documents, and conversations exist on a machine no other user can touch. No shared database. No lateral exposure.

Zero Trust
13 — Data Universe
120+ Sources. One Agent.
93
DBnomics Providers
1.7B
Economic Time Series
816K
FRED Series
30+
Direct API Integrations
250M
Academic Works (OpenAlex)
International Organizations
IMF World Bank OECD BIS ILO FAO UNCTAD UN Data UNDP UNESCO UNIDO WHO WTO CEPII GGDC Freedom House ND-GAIN
European Institutions & Central Banks
ECB Eurostat European Commission AMECO JRC ENTSO-E Bundesbank Banque de France Bank of England NBB (Belgium) DESTATIS INSEE ISTAT INE Spain ONS (UK) OBR (UK) SCB Sweden Statistics Poland CSO Ireland ELSTAT Greece Statistics Portugal SECO Switzerland TCMB Turkey Rosstat Russia
United States
Federal Reserve BEA BLS CBO EIA (Energy) FHFA (Housing) ISM Johns Hopkins NAR (Realtors) UMich Consumer Opportunity Insights
France
ACOSS DARES DREES DGDDI Customs Enedis IPP Météo-France Pôle Emploi RTE CEPREMAP
Asia-Pacific
Bank of Japan ESRI Japan JILPT METI Statistics Japan NBS China SAFE China Bank Indonesia MOSPI India RBA Australia
Americas
Banco Central do Brasil Bank of Canada Statistics Canada INDEC Argentina INEGI Mexico
Africa & Middle East
African Development Bank Africa Info Highway BCEAO (West Africa) Franc Zone SAIS-CARI SAMA Saudi Arabia SARB South Africa
Direct API Integrations — Real-Time
FMP — Financial Modeling Prep FRED — Federal Reserve Economic Data SEC EDGAR Polymarket Exa AI — 10B+ Pages VERIDIUM — 13.9K OSINT Docs Open-Meteo Weather Tavily (Real-Time Web) Firecrawl (Deep Crawl)
Research & Academic
arXiv OpenAlex — 250M+ Works PubMed — Biomedical PubChem — Chemistry UniProt — Proteins Supadata — YouTube
Productivity, Travel & Legal
Google Workspace Notion Duffel — 300+ Airlines Booking.com Matsne — Georgian Law Context7 — Framework Docs
Communication & Delivery
Telegram AgentMail ElevenLabs TTS GitHub Browser Use Spotify
14 — The App
Your Command Center

The DataDesk AI interface — where you interact with Mark-1 in real time. A three-panel workspace featuring conversation history, an intelligent chat interface with quick-action prompts, and a live system overview panel showing memory state, skill inventory, and agent status. Dark-themed, distraction-free, built for depth.

Mark-1 DataDesk AI App Interface
Left Panel
Full conversation history with search, organized chronologically. Every session persisted.
Center
Multimodal chat with file uploads, voice input, quick-action cards, and real-time streaming responses.
Right Panel
Live system overview — agent status, memory banks, Second Brain, 106 skills, and quick actions at a glance.

Share Mark-1

Share this intelligence profile with anyone via link or social platforms.