Changelog
What’s new in Sable.
New features, improvements and fixes — shipped every week. Follow along as we build.
v2.0.0
New
Evals in CI
Score every pull request against your golden datasets. Sable posts a diff of what improved and what regressed, and fails the build when quality drops below your gates.
New Sable CLI with GitHub, GitLab and Buildkite integrations
Configurable quality gates per eval suite
Side-by-side run diffs linked from PR comments
v1.14.0
New
Critical-path view for multi-agent runs
Multi-agent traces now open on the critical path. Parallel branches collapse into a single row, and handoffs show the exact context passed between agents.
Automatic critical-path detection
Collapsible parallel branches
Context inspector on every handoff edge
v1.13.0
Improved
Guardrails can pause and escalate
Guardrails can now pause a run mid-flight and route it to a reviewer in Slack or your ticketing tool, with the full trace attached.
New pause action alongside block and redact
Slack, Linear and Zendesk escalation targets
Reviewer decisions are saved as labeled eval examples
v1.12.2
Improved
Faster trace search
Full-text search across prompts and completions is now up to 6× faster on large projects, with filters for model, tool, cost and eval score.
p95 search latency under 400ms on 1B+ spans
Saved searches and shareable filter links
v1.12.1
Fixed
Token counts for streamed responses
Fixed an issue where token usage for streamed Anthropic and Gemini responses was under-reported when the stream was cancelled early.
Accurate usage on cancelled streams
Backfilled cost data for affected runs
v1.12.0
New
TypeScript SDK 1.0
The TypeScript SDK is now generally available, with first-class support for the Vercel AI SDK, LangGraph.js and edge runtimes.
Zero-dependency core under 9 KB gzipped
Works in Node, Bun, Deno and edge runtimes
Automatic instrumentation for the Vercel AI SDK