3.23.0 (2026-11-03)
Features
agent and evaluator versioning with deploy labels, plus experiment and dependency fixes (#9124) (ec774cb)
coding-agent: capture Grok CLI sessions through its native OpenTelemetry (#9224) (34c0f70)
coding-agent: track pi as a natively instrumented coding agent (#9203) (52eaa9e)
dataset: dataset version history with autosave, compare, restore, retention and run pinning (#9121) (f50c6f0)
insight: a board no longer pops the daily insights offer on open (#9242) (e762c3d)
model-provider: run Instant Evals on Jev, Azure, OpenRouter and Fireworks decision models (#9249) (98482a1)
Bug Fixes
analytics: reconvergence replaces an LWQL view that lacks a catalog column (#9246) (16bb190)
auth: leave the page when a reader signs out (#9210) (7998642)
codeblock: include runner stderr tail in runner error message (#9238) (02108a8)
codeblock: run code blocks on Lambda, where seccomp refuses PR_SET_DUMPABLE (#9240) (7ddff2e)
design-system: dialogs stay centred while Langy is docked (#9254) (f36b967)
docs-site: keep the changelog Markdown export within the page token budget (#9259) (49f512b)
email-domain auto-join respects the seat cap, and the SSO refusal says who to ask (#9226) (95d4da8)
governance: validate pull source config on save and let admins edit the Copilot Studio connection (#9279) (862bc0e)
langevals: detect person names written in capitals (#9213) (4c46055)
langevals: skip the capitalised-name second pass for lone capital words (#9256) (749e9f3)
lwql: probe the approved view in the app's PostgreSQL schema (#9264) (b3547c3)
model-provider: judge Instant Evals on the saved decision model and the project's own key (#9272) (1d4055e)
model-provider: offer and accept decision models per edited scope (#9294) (a5114a1), closes #9293