Most teams face a tradeoff: use many ad-hoc single-agent setups that leak secrets and diverge, or rely on hosted agent SaaS that places code, logs, and credentials outside your boundary. Centaur flips that tradeoff by offering a team-shared control plane that runs each conversation in an isolated Kubernetes sandbox, persists execution state, and confines real credentials to a network-edge proxy.
What Sets It Apart
- Per-conversation Kubernetes sandboxes with a default-deny network policy — so each Slack thread gets a real, reproducible runtime (shell, git, Python, Node, etc.) instead of a stateless LLM reply. This means agents can run tests, edit code, and interact with repos in a real environment.
- Credential-safe outbound access via an iron-proxy that injects secrets only at the network edge — so long-lived API keys are never exposed inside the sandbox. This reduces blast radius while allowing agents to call approved services.
- Durable workflows and replayable state — runs can sleep, resume, wait for events, spawn child agents, and survive restarts; transcripts and events are persisted for audit and reconnection.
- Shared tool/plugin model and harness-agnostic execution — add a Python tool once and make it available to every conversation; run different CLI harnesses (Amp, Claude Code, Codex, or custom) under the same orchestration.
Who It's For and Trade-offs
Great fit if you run teams that need agents to perform long-lived, instrumented tasks against internal codebases or services while keeping secrets inside your perimeter. It benefits orgs that want standardized agent behavior, centralized auditing, and reusable tool integrations. Look elsewhere if you only need a lightweight single-user assistant or cannot operate any Kubernetes-like infrastructure — Centaur requires self-hosting and operational ownership, and its shared credential vault and proxy introduce platform-level complexity and trust assumptions you must manage.
Where It Fits
Centaur positions itself between one-off local agent setups and hosted agent SaaS: it centralizes orchestration and governance without outsourcing data or keys. For teams evaluating safe agent adoption, it accelerates org-wide rollout by converting integrations and workflows into reusable primitives under a single control plane.
