Demos

Recorded against a running install — no mockups, no edits, nothing staged that the platform does not do. Start with Agent Galaxy: it shows a call succeeding, a permission changing, the next call failing, and the evidence for both.

Agent Galaxy

Traffic in real time

  • The galaxy pulses live traffic — human to agent to tool call to data securable.
  • The first prompt succeeds with two calls, and the full call chain is there as evidence.
  • Then permissions change. Deterministically the next call fails, and the audit trace says so.
  • From the trace the operator diagnoses and remediates. The subsequent call traces all the way to the Snowflake lakehouse and succeeds.
  • Last, the admin reveals the prompt alongside the verifiable, reproducible evidence object.

Catalog Universe

Diagnosing a partial failure

  • The same live graph, but this prompt only partially works — one tool call fails because a table carried no permission.
  • The audit log shows why: the mcp-consumer role had no select on that precise schema.
  • The admin grants traversal on the schema, and the display fills in the coverage the red bricks were marking.
  • The same prompt then succeeds, and the audit trace shows the completed chain.

Chargeback and reporting

Where the money went, by whoever asks

  • A set of reports on a monthly cadence, out of the box.
  • Chargeback statements for the current month or across months.
  • The aggregated total is the same figure across several of them — one record, several readings.
  • Totals by vendor, and by department, user and service account.

Registering an LLM

Provider, contract and rate card

  • Register a new LLM, choose a model, add the credentials.
  • Attach a cost contract with the known details — input, output, cache read and cache write, each with its own unit cost.
  • Set the spend limit here; other thresholds are editable later.
  • Review, register, then view the agent integration specifics — the same details your developers see in the developer portal.
  • When the review process has completed, approve the LLM for use.

Model comparison

Same prompt, every registered provider

  • How many LLM providers are registered, and how do they compare on an arbitrary prompt?
  • Each model returns its answer, and Magertron computes the timing, the cost, and the input and output tokens for each.