CHANGELOG

Only launch-approved changes.

This feed includes changes reviewed for Staso's current Observe and Guard launch. A historical release can contain work that is not generally available here.

For current availability, check Product and Docs.

SEPTEMBER 2026
  1. v0.12.0
    latest

    Safer setup and accurate trace usage

    • NEWTeam now starts at $39 USD per month for one unit.
    • NEWEach Team unit includes 10,000 traces and 5,000 Guard LLM-judge evaluations per month.
    • NEWStatic Guard evaluations are unlimited on Team.
    • IMPImproved setup profiles and credential repair for Claude Code and Codex.
    • IMPImproved Codex turn grouping, steering, and lifecycle handling.
    • IMPRemoved the per-trace span ceiling. Traces are no longer rejected for their span count.
    • FIXLinked each Guard finding to the exact offending span.
    • FIXCounted unique monthly traces so resends no longer inflate usage.
APRIL 2026
  1. v0.9.1

    Explicit agent traces across supported provider tool loops

    • NEWTool-loop state machine stitches multi-turn tool calls into a single trace for Anthropic and OpenAI
  2. v0.9.0

    Guard failure modes

    • NEWPick fail-open or fail-closed per Guard call, or default it process-wide
  3. v0.7.0

    Guard audit and enforce controls

    • BRKBreaking: `agent_id` renamed to `agent_name` across SDK and docs
    • NEWAudit mode records a would-block and lets the proposal continue. Enforce mode can block before normal dispatch.
    • NEWPolicies scope rules to specific agents, tools, or environments
    • IMPOpenAI traces render with provider-aware pretty cards