01
Autonomous Execution Engine
A multi-turn model ↔ tool execution loop powered by Google Gemini (gemini-2.5-flash), with persistent task checkpointing in SQLite.
Flagship AI Innovation · Production Live
OMENA Workbench— Autonomous Agentic IDE & DevOps Platform
A custom-engineered autonomous AI engineering distribution built on Open WebUI. Empowers autonomous AI agents to write, test, self-heal, and deploy complete applications inside a sandboxed Linux environment with zero human intervention.
178+ req/sec ThroughputZero-Trust Sandbox ConfinementSelf-Healing RuntimeMobile & Desktop IDE
From intention to execution
An engineering loop.
Not just a chat window.
Model reasoning meets real tools, a persistent workspace, and a runtime that can repair its own failures. Follow every task through streaming logs and durable checkpoints.
/workspace
app.py
requirements.txt
tests/
checkpoint.db
Build. Test. Repair. Deploy.
01 Plan task → persist checkpoint
02 Execute tools inside /workspace
03 Inspect stderr → diagnose → patch
04 Re-run tests → record task receipt
Confined execution Persistent progress
Gemini ↔ tools · SQLite checkpoints · streaming logs
Core capabilities
Autonomy, with engineering discipline.
02
Autonomous Self-Healing
Intercepts runtime stderr tracebacks, diagnoses bugs such as SQLite concurrency locks, and writes code patches on the fly without human intervention.
03
Workspace Boundary Sandboxing
Strict path-traversal confinement to /workspace, execution timeouts, and zero-secret environment variable sanitization keep agent execution contained.
04
3-Tier Policy Governance
Configurable SAFE read-only audit, ASSISTED semi-autonomous, and AUTONOMOUS full-development modes put execution scope under your control.
05
Agentic IDE Dock
An interactive browser terminal with streaming logs, a live workspace file explorer with inline editor, and real-time task receipts and progress monitoring.
06
Mobile Ergonomics
A touch-friendly overlay drawer with dark-backdrop dismissal, optimized for iOS and Android. Product-team testing includes iPhone screens at 390 × 844.
Execution boundaries
Three modes. One controlled workspace.
01 / SAFE
Read-only audit
Inspect the workspace and review findings without changing files.
02 / ASSISTED
Semi-autonomous
Keep a human in the loop while agents assist with engineering tasks.
03 / AUTONOMOUS
Full development
Let agents write, test, self-heal, and deploy within the confined workspace.
Technical architecture & benchmarks
Built for a real production runtime.
AWS EC2
Ubuntu 24.04 LTS
Caddy
TLS reverse proxy
FastAPI / Uvicorn
Application & execution API
SQLite WAL
Persistent task checkpoints
Phase 9 · Production benchmark
- 500
- Requests
- 178.21
- Requests / second
- 95%
- Success rate
- 2.80s
- Wall time
Automated self-healing observed during the benchmark.
Results supplied by the OMENA Workbench team from its verified Phase 9 run; not independently reproduced by this website. Throughput and success rates depend on workload and deployment conditions.
Transparent distribution
Independent innovation.
Upstream respect.
OMENA Workbench is an independently engineered derivative distribution built upon upstream Open WebUI v0.11.4, with substantial enterprise agentic additions. It is not the official upstream Open WebUI product.
The distribution respects the Open WebUI License and retains upstream author attributions. Open WebUI remains credited to its original authors; the applicable upstream license and notices govern the underlying software.
OMENA Workbench