Conduit v0.16.0 release

Conduit v0.16.0 is the first release built end-to-end against the Phase 1 execution plan: a real first-five-minutes experience, structured errors instead of opaque strings, and the deployment basics — readiness, liveness, metrics, full env-config — that make Conduit boring to run in a container or a pod.

tip

Deterministic CLI exit codes, connector-config redaction in logs, and a CI guard enforcing error-code coverage landed on main shortly after and ship in the upcoming v0.16.1.

Key Highlights

• New conduit quickstart command: one line scaffolds and runs a demo generator → log pipeline in-process — records flow to your console within seconds, no flags, nothing written to your working directory. It's the same generator-to-log combo from our pipeline guide, now zero-config.
• Structured errors (ConduitError): user-facing errors are gaining a stable machine-actionable shape — code, the failing config path, and a suggested fix — surfaced over the gRPC/HTTP API so agents and the MCP server can act on an error instead of just printing it. Rollout is incremental across error sites.
• --json now works on every read command (list/describe) across connectors, connector plugins, processors, processor plugins, and pipelines, not just pipelines list. See the CLI reference.
• 12-factor essentials: a new /readyz readiness endpoint joins the existing /healthz liveness check and Prometheus /metrics — /readyz returns 503 while the engine is starting or its state store is unreachable, and reports degraded pipelines in the response body without flipping the process to not-ready. Every setting is env-configurable (CONDUIT_*, flag > env > file precedence), and conduit run --pipelines <dir> is now a shorthand for --pipelines.path.
• Lifecycle correctness: a degraded pipeline now reports the source connector's actual error instead of a downstream io.EOF that masked it, and force-stop / WaitPipeline no longer risk a nil-pointer panic or a lost terminal error under race conditions during startup and shutdown.