orbitaven 0.3.0

Autonomous Orchestration Platform — Signal. Simulate. Execute.

Orbitaven

Autonomous Orchestration Platform — Signal. Simulate. Execute.

Orbitaven is a local-first, profile-driven orchestration platform that watches for signals, builds workflows, assigns agents, simulates execution, and controls digital and physical systems — all through a real-time chat interface. Runs 24/7 on your own Ubuntu hardware with zero cloud dependency.

---

What it does

• Listens — monitors URLs, RSS feeds, APIs for trigger events 24/7
• Thinks — classifies intent, generates multi-step workflows via LLM
• Acts — dispatches real agents (file, web, shell, code, data, memory…)
• Remembers — persistent per-profile memory and conversation history
• Grows — creates new agents through chat conversation
• Shows — real-time streaming UI with workflow panel, signal alerts, agent pool

---

Architecture

orbitaven/ ├── core/ │ ├── config.py — env-based settings │ ├── database.py — SQLite WAL, all tables │ ├── auth.py — JWT tokens │ ├── profile.py — profile CRUD, memory, history │ ├── llm_adapter.py — unified LLM interface │ ├── brain.py — intent classifier + workflow generator │ ├── workflow_manager.py — state machine (pause/resume/restart/cancel) │ ├── agent_registry.py — agent pool (10 built-in + generated) │ ├── agent_factory.py — LLM-powered agent generator + wizard │ ├── credential_vault.py — encrypted API key storage (Fernet) │ └── signal_monitor.py — 24/7 background signal daemon │ ├── agents/ │ ├── base_agent.py — abstract base class │ ├── file_agent.py — local filesystem operations │ ├── web_agent.py — URL fetch + DuckDuckGo search │ ├── shell_agent.py — safe whitelisted shell commands │ ├── memory_agent.py — persistent profile memory │ ├── summarizer_agent.py — LLM-powered content condensing │ ├── coder_agent.py — code generation + sandboxed execution │ ├── data_agent.py — CSV/JSON parsing + statistics │ ├── notify_agent.py — alerts + notification log │ ├── signal_agent.py — URL/RSS/API monitoring │ └── generated/ — auto-created agents via chat │ ├── api/ │ ├── main.py — FastAPI entry point │ ├── websocket.py — streaming WebSocket chat handler │ └── routes/ │ ├── auth.py — profile login, JWT, LLM config │ ├── workflow.py — workflow CRUD + control │ ├── agents.py — agent pool endpoints │ ├── signals.py — signal CRUD + manual check │ ├── factory.py — agent generation + vault CRUD │ └── system.py — health, stats, platform info │ ├── ui/ │ └── index.html — monochrome sci-fi single-page app │ ├── data/ │ ├── orbitaven.db — SQLite database │ └── profiles/ — per-profile vault + generated agents │ ├── run.sh — single command startup ├── requirements.txt └── pyproject.toml

---

Build Phases

Phase	Scope	Status
1	Core foundation — config, DB, profiles, LLM adapter, WebSocket chat, UI	✅ Done
2	Brain engine — intent routing, workflow generator, pause/resume/restart	✅ Done
3	Agent pool — 10 built-in agents, registry, real execution in workflows	✅ Done
4	Signal system — 24/7 monitor, rule engine, alert UI, one-click launch	✅ Done
5	Agent factory — self-growing agents via chat, encrypted credential vault	✅ Done
6	Polish — health dashboard, system stats, run.sh, README	✅ Done
7	Public Multi-Agent Negotiation Protocol — external agent collaboration	🔜 Planned

---

Quick Start

# 1. Clone
git clone https://github.com/pinaki/orbitaven.git
cd orbitaven

# 2. Create virtualenv
python3 -m venv .venv
source .venv/bin/activate

# 3. Install
pip install -r requirements.txt

# 4. Configure
cp .env.example .env
# edit .env if needed

# 5. Run
./run.sh

Open browser at http://localhost:8000

---

LLM Configuration

Each profile configures its own LLM. Supported providers:

Provider	How
ollama	Local Ollama — http://localhost:11434
kaggle_tunnel	Kaggle ngrok/cloudflare tunnel URL
openai	OpenAI API key
anthropic	Anthropic API key
openai_compatible	LM Studio, Groq, Together, etc.

Switch model anytime from Settings panel or chat.

---

Built-in Agents

Agent	Capability
assistant	General LLM reasoning, writing, answering
file_agent	Read, write, list, search local files
web_agent	Browse URLs, DuckDuckGo search
shell_agent	Safe whitelisted shell commands
memory_agent	Persistent per-profile memory store
summarizer_agent	LLM-powered content condensing
coder_agent	Code generation + sandboxed Python execution
data_agent	CSV/JSON parsing, calculations, statistics
notify_agent	Alerts, notification log
signal_agent	URL/RSS/API monitoring

---

Self-Growing Agents

Type in chat: create an agent for GitHub

Orbitaven guides you through:

1. Name + description
2. Capabilities
3. API keys (stored encrypted)
4. Generates Python agent file
5. Registers it in the agent pool immediately

---

Signal System

Configure signal sources per profile:

• URL watch — keyword or status change detection
• RSS feed — new item detection
• API watch — threshold-based triggers (gt/lt/eq/contains)

When triggered: alert appears in UI → one click to launch a suggested workflow.

---

API Endpoints

Auth GET/POST /api/auth/profiles, /login, /me, /llm Workflows GET/POST /api/workflows, /{id}, /{id}/pause|resume|restart|cancel Agents GET/POST /api/agents, /{name}/run Signals GET/POST /api/signals, /{id}/check, /alerts Factory GET/POST /api/factory/create, /agents, /vault System GET /api/system/health, /stats, /info WebSocket WS /ws/chat

---

Hardware Requirements

Minimum (orchestration only):

• Ubuntu 20.04+
• 4GB RAM, 10GB disk
• Python 3.11+
• No GPU needed

For local LLM inference:

• Ollama handles separately
• Or use Kaggle tunnel (free GPU on Kaggle)

---

Phase 7 — Planned

Public Multi-Agent Negotiation Protocol

External developers register agents via API key. Agents can:

• Share requirements (buyer agent)
• Respond with offers (vendor agents)
• Negotiate terms (no information leakage between agents)
• Root orchestrator evaluates and decides

Open protocol — any agent can participate. Documentation and SDK planned post-Phase 6.

---

Tech Stack

Layer	Technology
Backend	Python 3.12 + FastAPI
Database	SQLite (aiosqlite) — WAL mode
Realtime	WebSocket streaming
Auth	JWT (python-jose)
Encryption	Fernet (cryptography)
LLM	Any via unified adapter
Scheduler	asyncio + APScheduler
UI	Vanilla HTML/CSS/JS — no build step

---

License

MIT © 2026 Orbitaven