Ryph

Quick answers

How do agents handle incident response?

When an alert fires, an agent joins your team's incident chat (Slack, Discord, Telegram or Zulip). It reads the alert, queries your metrics and logs, reviews recent commits and runbooks, and posts what it finds. Team members can mention it to steer it. Once it identifies the cause, it can write a fix, run your checks, and open a pull request for review. When the alert resolves, it wraps up the conversation.

What prevents agents from being exploited with hostile input?

Ryph assumes any agent can be manipulated by hostile issues, web pages or chat messages. Credentials stay behind gateways, agents have no git keys, and work leaves the sandbox as patches that require review. All tool calls go through allowlists, internet requests through an egress proxy with host restrictions, and every request is logged and checked against what the agent is allowed to do.

How does Ryph's semantic memory work?

Every agent run adds to a shared memory. Ryph uses semantic search to find duplicates and contradictions, consolidating multiple rephrased claims into one confirmed fact. If code changes invalidate a memory, Ryph flags it so it's rechecked on next use. Confirmations from multiple runs help rank memories by relevance, age and trust.

What deployment options does Ryph support?

Ryph runs as a single binary on your laptop, with Docker Compose on a server, or with Kubernetes for team deployments. Environment builds are on by default for the binary and Docker Compose, and on Kubernetes an operator enables them. In all cases, your code stays on your own infrastructure.