Warren turns coding agents into infrastructure
Warren is an MIT-licensed, self-hostable control plane for running coding agents in isolated, observable workloads. It supports sandboxed execution, live events, spend limits, recovery, and Git branch or pull-request delivery.
Warren targets the real bottleneck in autonomous coding: operational control around agent runs, not another agent interface. Its shared-token model still limits it to solo operators and trusted teams, but the infrastructure-first approach is compelling.
- –Supports local, Docker, and Kubernetes runtime backends with per-run isolation.
- –Unifies UI, CLI, and HTTP API workflows around one observable run lifecycle.
- –Adds steering, cancellation, recovery, spend caps, concurrency limits, and scheduled runs.
- –Harness adapters let teams operate Claude Code, Pi, and other agents through the same control plane.
- –Self-hosting keeps code, credentials, compute, and run history under the operator’s control.
DISCOVERED
1d ago
2026-08-26
PUBLISHED
1d ago
2026-08-26
RELEVANCE
AUTHOR
Jaymin West