Coding Agent v2: Iterative Deployment
A second-generation coding agent that closes the full deployment loop — it can identify runtime failures, propose fixes, rebuild, and re-release autonomously. Includes LLM selection for more capable reasoning, with token fees applicable.
Coding Agent v2 adds an iteration loop to the first-generation agent. Where v1 can deploy an SDL config, v2 can watch the deployment, detect failures in logs or health checks, propose a fix, and re-deploy — autonomously.
The agent understands common failure modes: image pull errors, OOM kills, port binding conflicts, crashing containers, and failed health checks. For each, it applies a known resolution strategy or escalates to the developer with a specific recommendation.
The iteration loop is bounded: the agent operates within a configurable number of retry cycles and always presents a summary of what changed between attempts. Destructive actions — scaling to zero, deleting a deployment, replacing a volume — always require explicit developer approval.
v2 also introduces LLM selection: developers can choose the model powering the agent — from lighter, faster models for routine iteration to frontier models for complex debugging and architectural reasoning. Token consumption is metered and billed per use.