Tale v0.5.91

Highlights

• Temporary provider-capacity refusals no longer trip the hourly automated-start circuit. Fleet Manager keeps dispatching through short OpenAI capacity pressure, while each fresh delegated or scheduled start still observes the provider's one-minute retry floor before trying again (#4985, #4945). Real automated starts continue to count toward the circuit, so repeated task failures still receive the existing protection.

Upgrade notes

• From v0.5.90 to v0.5.91 no database migration or configuration change is needed, and the API contract stays at 3.44.0. The fleet scheduler adopts the retry behavior on its next run after the backend is upgraded; no manual task or schedule changes are required.

API contract changes

None in this range. The contract stays at 3.44.0: 161 operations.

What's Changed

• fix(platform): log why an agent's metadata document fetch failed by @larryro in #4982
• feat(sandbox): run sessions on Incus by @yannickmonney in #4814
• feat(cli): read a deployment spec written as YAML by @yannickmonney in #4983
• fix(platform): keep capacity refusals out of task circuit by @yannickmonney in #4985
• docs: prepare Tale v0.5.91 notes by @yannickmonney in #4986

Full Changelog: v0.5.90...v0.5.91