ANOMX AGENT
In the platform. In the background. In your terminal. An AI partner that connects data and tools to the context of your machines.
WHERE YOU WORK
Ask questions in a shared workspace. Investigate systems, channels, jobs and findings together, with connected tools and your team’s context.
Turn a question into a recurring assignment. Planned prompts investigate change, revisit earlier runs and preserve findings as recommendations.
Work where data and code live. The open-source CLI agent inspects local files, runs permitted tools and connects to the platform when needed.
THE RIGHT DEGREE OF AUTONOMY
Planned prompts run in the platform without questions. The evaluator makes approval decisions within existing permissions. By default, the agent reads and creates recommendations. Only explicitly allowed object types can be changed directly. Budgets and previous runs define the context.
Reads and authorized, reversible work proceed without interruption. The agent asks only about consequential risks or material changes to your request. While you decide, it can continue independent work.
The default for new sessions. The evaluator considers your request, risk, and earlier decisions, then allows or denies actions on your behalf. You receive no approval requests.
The agent skips command evaluation and executes its actions without approval requests. Existing access permissions and technical limits of the execution environment still apply.
Three interactive modes support direct work; in the CLI, cycle with Shift+Tab. Background is a separate platform mode for planned prompts and is not part of that cycle.
FROM A PROMPT TO AN ONGOING ASSIGNMENT
Define what the agent should investigate and when it should return. It connects new observations to earlier runs and keeps its work inspectable.
Data discovery at DESYPlanned prompts define a goal and interval. The platform coordinates execution.
Connected platform objects, API tools and skills make systems available for investigation.
Recommendations, run history and usage budgets make the work reviewable. Direct changes require an explicit object scope.
ALSO AT HOME IN YOUR TERMINAL
A standalone starting point for code, data and anomaly analysis. Start Anomx in your working directory and connect your model provider.
pip install anomxanomx$ anomx Anomx. › Investigate these sensor readings. What changed? Inspect the data structure Compare channels and patterns Explain candidate anomalies Λ Auto Approve · Shift+Tab
DOCKER & PODMAN
Optionally run agent commands in a container. Anomx prepares the runtime and preserves your task data. Specific local project folders can be mounted with read/write access. Your selected mode determines whether you, the evaluator, or the agent decides on access.
Set up sandboxingOPEN TO YOUR MODELS
Connect a cloud provider or a locally served model. Model access, tools and the selected mode determine the available capabilities.
THE NEXT ERA
Build the future of autonomous systems with us.
Request early access