Skip to main content

Season 1: Trust + Capability

Theme​

Make Seraph's guardian promise operationally credible.

Why this season is first​

Seraph already has a differentiated thesis, but it still trails the best competitors on runtime credibility. This season fixes that before we invest further in reach or delight.

Win condition​

By the end of this season, Seraph should be safer, broader in execution, and more reliable under real use, with enough guardrails that future presence work is worth shipping.

Batches​

S1-B1 Trust Boundaries​

  • policy profiles
  • approvals for sensitive actions
  • secret scoping
  • auditable action flow
  • stronger isolation model

S1-B2 Execution Plane​

  • real shell and process execution
  • stronger browser automation
  • workflow engine direction
  • a clearer path from “reasoning” to “doing”

S1-B3 Runtime Reliability​

  • model routing and fallback
  • local model path
  • observability
  • evaluation harness
  • predictable degraded-mode behavior

Exit criteria​

  • security and trust boundaries are materially stronger than today
  • the execution plane no longer feels toy-sized next to the product thesis
  • the runtime can fail more gracefully and more observably
  • implementation work can proceed without constantly reopening core trust concerns