See Every Step
Your AI Takes
Traces tell you a model was called. Prometheux shows why the answer came out the way it did: each agent step as it runs, exactly what every LLM call was given, and the rules and rows behind any result.
Proven in production
Logs Show Calls.
We Show Reasons.
Tracing tools record that a model was called, how long it took and what it cost. To trust an AI system you also need what it was given, what it did and which logic produced the result.
Watch the Agent
Work
The Prometheux AI Assistant builds ontologies, writes and runs concepts and inspects data in the open. You see each step as it happens, and every conversation stays in your chat history.
Live Step Stream
Replies, running tools, proposed changes and step results stream as the assistant works, not after the fact.
Chat History
Past conversations are kept and auto-titled in Chat history, where you can rename or delete them, so you can revisit what was done.
Confirm Before Damage
The in-app assistant asks before making changes and always before deleting anything. Our MCP tools are labelled read-only or destructive, so your MCP client can ask first.
Artefacts, Not Chat
What the agent builds is saved as concepts and rules, which you can inspect and rerun.
Know what your agents did, not just what they said. Catch risky actions before they happen.
Every Run Says
What It Was Given
LLM concepts are first-class steps in the ontology. Each run reports its inputs and limits, and deterministic results carry provenance you can materialise and query.
Input Accounting
Each LLM concept run reports how much upstream input was included and whether the prompt was trimmed to fit.
Cut-Off Replies Flagged
If a reply is cut short by max_tokens, the run says so, instead of passing a partial answer downstream.
Usage on Demand
Query current usage, limits and remaining quota for LLM and embedding calls through the REST API.
Queryable Provenance
Materialise the chase graph to CSV or Neo4j and query exactly which facts and rules produced each result.
Spot silent failures before they reach users. Track AI usage without bolting on another tool.
From Black Box
to Glass Box
Model the Work
Business logic and LLM steps become named concepts instead of hidden prompts.
Run With Reporting
Each run reports its inputs, trimming and truncation; the assistant streams every step.
Inspect the Result
Open the rules and rows behind any deterministic result through provenance and explanations.
Track Usage
Check LLM and embedding usage against your limits via the API.
Visibility Where
AI Can Do Damage
Agents in Critical Workflows
Teams see each action an agent takes, and agents act with the same access rights as the user they work for, before AI touches anything operational.
Explaining Model-Assisted Output
When an answer combines rules and an LLM step, reviewers can see which part was derived and which was generated.
Usage Across Teams
Platform owners read LLM and embedding usage and remaining quota programmatically.
Debugging Results
When a number looks wrong, provenance shows which rule and which rows produced it.
Chase & provenance →Questions,
Answered
What does AI observability mean in Prometheux?
Seeing what AI steps did, what they were given and which logic produced each result. The AI Assistant streams its steps, LLM concepts report their inputs and truncation, and deterministic results carry provenance.
Does this replace our LLM tracing tool?
No, keep it. Tracing tools log calls, latency, cost and evaluation scores across your AI stack. Prometheux makes the logic itself inspectable: business rules are named concepts, LLM steps report their inputs and truncation, and deterministic results carry provenance down to the source rows.
Can I see when a prompt was truncated?
Yes. Truncation is never silent: each LLM concept run reports how much input was included, whether the prompt was trimmed and whether the reply was cut short by max_tokens.
Does our data leave our environment?
It doesn't have to. Prometheux can run in your own cloud, inside Snowflake or Databricks, or on-premises, and reads data where it lives with no copies, under your existing access controls. All queries and access are logged.
Can I track LLM usage?
Yes. The REST endpoint GET /api/v1/users/usage-status returns current usage, limits and remaining quota for LLM and embedding calls.
Related Capabilities
Stop Guessing What
Your AI Did
Bring an AI workflow you can't see inside. We'll show you every step, input and rule behind it.
