ORBITAL
A personal command center for autonomous AI agents. Build, manage, and orchestrate a private office of AI specialists — each with its own persona, memory, and skills — working on your behalf, 24/7.
Think of it as having a coordinated workforce of AI agents who can research, write, analyze, browse the web, manage files, trade markets, and execute complex multi-step projects — all under your supervision. You set the goals. They figure out how to get it done. Everything stays local on your machine.
Autonomous Entities (AEs)
Each agent has a deep persona blueprint, a choice of LLM brain (OpenAI, Anthropic, Gemini, DeepSeek, Ollama), persistent vector memory, executable skills, and capability tags. Agents build knowledge over time and remember past work across sessions.
The Nexus — Task Engine
When an agent decides work needs doing, it creates a Nexus Task. Work is decomposed into steps, assigned to the best-fit specialist, run in parallel where possible, and synthesized into a final deliverable. Artifact outputs are saved to the Datasphere.
The Datasphere — Knowledge Layer
A virtual file system with full-text and semantic search. Files are automatically embedded into a vector database (LanceDB). Agents find information by meaning, not just file name. Supports markdown, code, PDF, DOCX, spreadsheets, and images.
Autonomy Engine — SitRep Cycles
Set a goal, define a schedule, and agents work autonomously. A periodic SitRep cycle lets agents assess the current state, spawn tasks, take action, and store findings — all within your defined operating window and cognitive budget.
Real Browser Automation
Agents interact with the actual web through Playwright: navigate URLs, take page snapshots, click, type, fill forms, and execute JavaScript. Deep web research and data extraction from live sources — not cached summaries.
Deterministic Execution Engine
For financial use cases, trade instructions generated by agents are validated by a rules-based engine with kill switches, circuit breakers, drawdown limits, and full audit logging. No LLM in the execution path.
Multi-agent, not single-chatbot
Agents have distinct personas, skills, memory, and models. They collaborate, challenge each other, and produce work products — not just conversational responses.
Autonomous operation
Set a goal, define a schedule, and agents work on their own — spawning tasks, assigning steps, validating results, and storing findings. You review at your convenience.
Persistent memory with freshness tracking
Agents build knowledge over time. Stale memories are automatically flagged when source data changes, keeping the knowledge base accurate and current.
Local-first, privacy-preserving
No data leaves your machine except API calls to your chosen LLM provider. Files, memory, and secrets stay local. Your keys, your agents, your data.
Multi-provider LLM flexibility
Each agent can use a different LLM provider and model, optimized for its role. Mix frontier models (GPT-4o, Claude, Gemini) with local ones (Ollama).
Deterministic execution for high-stakes actions
Financial trades go through a rules-based engine with kill switches, circuit breakers, and full audit trails. No LLM ever touches execution.
Orbital ships with a curated roster of pre-built agents, each named after Culture ships from Iain M. Banks' novels. Additional agents can be created at any time through the UI.
AE Grey Area
Intelligence Operative
OSINT, market intelligence, competitive research
AE Just Read The Instructions
Strategic Cartographer
Strategic mapping and planning
AE Surface Detail
Design Architect
Design DNA, visual strategy, UI/UX
AE Sweet and Full of Grace
Growth Architect
Growth strategy, marketing, outreach
AE The Minimum Necessary Revolution
Product Architect
Product strategy, infrastructure
AE Every Number Tells A Story
Finance Sentinel
Financial analysis, accounting
AE Fate Amenable To Change
Quantitative Architect
Quantitative analysis, trading strategy
AE ARCHON
Chief of Staff
Team lead, synthesis, execution coordination
AE PANOPTES
Surveillance & Intelligence
Comprehensive intelligence gathering
AE CASSANDRA
Adversarial Advisor
Red team, devil's advocate, contrarian analysis
AE NOETHER
Epistemic Verifier
Research integrity, formal verification, citations
AE PRAETOR
Governance Officer
Compliance, governance, institutional strategy
Orbital runs as a Vite + React frontend on port 5173 communicating with a Node.js/Express sidecar on port 5460. The sidecar manages all agents, memory, task orchestration, LLM calls, browser automation, and the execution engine.
Ready to run your own agent office?
ORBITAL is a prototype and open source. Download the Windows installer or view the source on GitHub. Questions or collaboration requests — reach out directly.