Changelog

What’s New

Evals for Tasks and Agents Test changes before you ship them with evals. Replay real past runs against a candidate revision and let a judge model score each result as pass or fail, so you catch regressions before they reach production. Eval pages are now shared across both Tasks and agents.

Agent Sharing and Cloning Share an agent with a public read-only link that shows its configuration without exposing credentials or run history, and let others clone it into their own project. Cloning is recursive and reports exactly what was copied, reused, or left for setup. Connections are copied as metadata only, so every project keeps control of its own credentials.

Memory Compaction Keep long-running agent conversations within the model’s context window using memory compaction. Agents can compact conversation history on a set cadence or automatically once a token threshold is reached, and compaction is billed and reported separately so you can see its cost.

Agents-First Navigation Agents are now the landing page after sign-in and sit at the top of the sidebar. A new Connect tab brings Slack channels, schedules, and API keys together for wiring up how an agent runs.

Improvements

• Support for the Standard Webhooks signature format when verifying incoming events
• New “created by me” filter on Tasks and revision and status filters on Logs
• Removed team member seat limits from the invite flow
• Added an OCR fallback so scanned and image-based PDFs are read more reliably
• Agents now handle context window overflow gracefully instead of failing the run
• Real Slack channel picker and a live next-run preview for scheduled agents