getomnia.com

What It Is

Omnia AI is positioned as an observability tool for AI Agents and is currently in private beta. Its core value proposition is to “see what the agent actually did”: by wrapping an existing agent run function with a decorator, it abstracts a run into a task with a goal, an agent graph, and result scoring, enabling end-to-end tracing from the original goal to the final output.

Core Capabilities

Based on the page content, Omnia focuses less on traditional LLM call logs and more on task-scope tracing. Each agent run becomes a task that records the goal, participating agents, tool calls, decision process, and final result. The native agent graph can visualize the structural topology of multi-agent runs, such as who created whom, what each agent said, and at which step an error occurred. Goal scoring uses asynchronous LLM-as-judge evaluation to score task results against the original goal, helping identify cases where “the workflow finished, but the goal was not achieved.”

Pricing and Trial

Currently, only a waitlist is available. The page states that users who join will receive early access when it opens, with a “no spam” note. Formal pricing, free quotas, trial duration, team plans, and enterprise plans have not been disclosed, so its cost structure and value for money cannot yet be assessed.

Pros and Cons

The main advantage is its clearly defined product focus: it targets task chains and multi-agent collaboration issues, which are among the hardest problems to debug in agent applications. The integration example suggests that developers only need to wrap a function with @observe(goal=...), making the developer experience look relatively lightweight. The drawbacks are also clear: it is still in private beta and public information is limited. It does not specify which agent frameworks are supported, what language SDKs are available, how data is stored, how access control is handled, or what the data retention policy is. The scoring model, accuracy, and configurability of the LLM-as-judge component are also undisclosed.

Who It’s For

Omnia is better suited for engineering teams building AI Agents, multi-agent workflows, or tool-calling applications, where it can be used for debugging, post-run review, anomaly investigation, and goal achievement evaluation. If you are only making simple calls to chat models, its agent graph capabilities may not be useful for now.

China Access

Access from mainland China, network stability, and payment methods have not been disclosed, so china_access can only be marked as unknown. If stable access is not available later, alternatives worth watching include LangSmith, Langfuse, Helicone, Arize Phoenix, or observability solutions based on OpenTelemetry.