Reliability GA + composition
Reliability became a feature of the engine, not a plan tier — and bots gained the primitives to compose.
Real automation fails sometimes; the engine now handles that as a first-class concern, on every account.
Reliability
- Try/Catch promoted to GA, with retry-before-catch (retry_count, retry_delay_ms) so a guarded branch retries before the catch runs.
- Per-step retries (retryOnFail, maxTries — default 3) on any node.
- Dedicated error outputs with on-error routing: stop the bot, continue on the regular output, or continue down a dedicated error branch.
- Full run history — workflow_runs and workflow_step_runs capture per-step input, output, status, timing and errors.
- Replay a run from any point with pinned or modified data, plus a run-level auto-replay policy for transient failures.
Composition
- Nested bots — call one bot as a step inside another (logic.execute_workflow).
- Loops with a fan-out toggle.
Known limit: loop fan-out currently executes downstream edges sequentially in edge order. True parallel concurrency is on the roadmap.