LoginSign Up
← Back to all sparks
S

Superwise

ANALYTICS
Velocity3.8

Machine learning model observability and monitoring platform

superwise.ai ↗

Superwise is turning its LLM gateway into a FinOps console for model fleets

llm-gatewayfinopsobservabilityguardrailsmodel-managementai-governance
◆Current state
Superwise's Sentinel line has gained an observability layer in two steps: a live fleet-wide activity log of every interaction crossing the gateways, then a FinOps dashboard reporting token consumption, cache-hit rate, reasoning ratio and output-to-input ratio by provider, model and sentinel. Both are explicitly metadata-only, with prompts and responses staying inside the customer's environment.
◆Where it's heading
The product is moving from policy enforcement toward cost accountability, and the metadata-only boundary is being used as the selling point rather than a limitation. Underneath, routine work continues on gateway sharing, policy previews and keeping the supported model list current.
◆Prediction
Cost visibility usually precedes cost control, so expect budget thresholds or routing policies that act on the FinOps metrics rather than only reporting them.

◆Recent moves

  1. 23d ago

    Sentinel FinOps dashboard tracks token spend across the fleet

    ⚡ SPARK

    A Sentinel FinOps dashboard reports token consumption and efficiency across the fleet — cache-hit rate, reasoning ratio, output-to-input ratio — broken down by provider, model and sentinel, with export. Turns a policy gateway into a cost-accounting surface.

    View source ↗
  2. 1mo ago

    Live fleet-wide activity log for gateway traffic

    A Sentinel Activity log streams every interaction passing through the gateways with provider, model, outcome and latency, metadata only. The observability groundwork the FinOps dashboard builds on.

    View source ↗
  3. 1mo ago

    Gateway sharing, policy preview, and CLI cleanup

    Gateways become easier to share, policies can be previewed before going live, and the CLI is smoothed out. Operational polish across the deployment path.

    View source ↗
  4. 2mo ago

    Explore gains new chart widgets; older models deprecated

    Opus 4.0, Sonnet 4.0 and Opus 4.1 are deprecated, a set of visualization widgets lands in the Explore experience, and a report-an-issue flow triggers automatic capture.

    View source ↗
  5. 2mo ago

    Top-K/Top-P config fix and two new Claude models

    A hotfix correcting Top-K and Top-P handling in LLM configuration, plus support for Claude Opus 4.7 and 4.8. Keeping the supported model list current.

    View source ↗
  6. 2mo ago

    Version 1.33.0

    The release notes carry no description of what changed. Nothing here is classifiable as user-visible work.

    View source ↗