Debug AI agents like you debug code.
NodeXLoop gives developers a visual, versionable environment for building agent graphs, stepping through execution, comparing models, and promoting proven workflows.
See the loop, not just the final answer.
Agent failures live across prompts, tools, memory, branches, and models. NodeXLoop makes the execution path inspectable.
Build visually
Connect agents, tools, gates, inputs, outputs, and human approvals in one graph.
Replay precisely
Return to the step that failed, inspect state, patch it, and prove the repair with a comparison run.
Promote deliberately
Pin model and workflow versions before moving a tested graph into AXIOM Runtime.
From graph to production instance.
The workflow remains portable and model-neutral while the production authority is applied at deployment.
Bring the models and tools you already use.
NodeXLoop does not require an Orivael model. Register an endpoint, define the rate, connect the workflow, and retain the graph as a git-native artifact.
- OpenAI-compatible endpoints
- NVIDIA NIM and open-weight models
- Local vLLM, Ollama, and custom gateways
- GitHub, Slack, Drive, APIs, and internal tools
| Artifact | Purpose |
|---|---|
| .loop.yaml | Portable, reviewable workflow definition |
| Run replay | Inputs, outputs, tools, state, and cost |
| Model registry | Endpoint, version, rate, and credential reference |
| Promotion manifest | Production version submitted to AXIOM Runtime |
Build the workflow before you buy the fleet.
Install NodeXLoop or bring one production candidate into an Orivael pilot.