Open-source incident response
for SRE teams
Monitor, diagnose, and fix safely — self-host free or run on our hosted Pro. Grounded in tool evidence. Safe by default.
agent@worker:~$ kubectl get pods -n production --field-selector=status.phase=Failed
- 98.5%
- fix success rate
- 3
- cloud providers
- 6
- CI/CD platforms
- 32
- MCP tools exposed
Deploy your way
Open source and self-hostable. Or let us run it.
Same agent, same safety rules — zero ops on Hosted Pro. Upgrade when you want a dashboard, email support, and managed infrastructure.
Self-host (Free)
MIT licensed. Full agent on your infra. Community support on GitHub.
Hosted Pro
We operate it. Incident dashboard, email support, monthly incident cap.
Used by teams evaluating autonomous incident response — design partner quotes coming as we onboard early customers.
Built for SRE & DevOps teams
A 24/7 teammate that handles the toil
Reduce alert fatigue
Let AI handle repetitive incidents — CrashLoopBackOff, OOMKilled, pipeline failures, gateway timeouts.
Faster MTTR
Automated diagnosis and remediation in minutes instead of a paging chain.
Grounded AI
Requires tool evidence before claiming a fix — reduces hallucinated remediation.
Durable audit trail
Org-scoped logs in S3, MinIO, GCS, or Azure Blob — every decision recorded.
Centralized or co-located
One agent server fixes remote EC2, Kubernetes, and cloud resources via SSH and native APIs.
Auto-escalation
Creates Jira or Zoho tickets and pings Slack/email when the agent can't resolve an incident.
Architecture
From alert to verified fix
Webhooks return instantly; a background worker diagnoses, executes safely, and documents every incident.
Central Agent Server (FastAPI)
Cloud Storage
S3 · MinIO · GCS · Azure
Org Docs + Audit
Checkpoints per org
Notifications
Slack · Email · Jira · Zoho
Remote targets — EC2 (SSH) · EKS/GKE/AKS (API) · AWS/GCP/Azure (API)
Safety first
Never delete. Never destroy. Notify instead.
Every destructive operation is permanently blocked. Every decision is logged. Approval gates keep a human in the loop.
Will it be hacked?
API key + webhook signatures + IP whitelist. TLS everywhere. Deploy behind a VPN or bastion. Regular security scanning in CI/CD.
Will it delete anything?
No. Destructive operations are permanently blocked. Email sent before any risky action, with manual approval required for anything critical.
Will fixes be accurate?
98.5% success rate with multi-stage verification, confidence scoring, and automatic rollback on failure.
Why DeAgent
Auditable, grounded, and managed when you need it
Pattern comparison — not vendor bashing. See how we stack up against typical alternatives.
| Feature | Black-box SaaS | Self-host scripts | DeAgent |
|---|---|---|---|
| Auditable | No | Yes | Yes |
| AI + tool evidence | Varies | No | Yes |
| Managed option | Yes | No | Pro |
| Grounded remediation | Varies | No | Yes |
| MIT / self-host free | No | Yes | Yes |
Platform support
Wherever your infrastructure runs
MCP server & BYOK
Bring your own agent. Bring your own keys.
Connect Cursor, Claude Desktop, or any MCP client to the DevOps toolkit. Each org keeps their own Anthropic, Slack, GitHub, and cloud credentials — never shared with other tenants.
↓ MCP (stdio or HTTP)
↓ org-scoped credentials
| Each org | Anthropic, Slack, GitHub, kube, cloud keys; their own AI agent |
| Platform operator | Shared storage/queue only — never org secrets |
// mcp-config.json
{
"mcpServers": {
"devops-ai-agent": {
"command": "devops-agent",
"args": ["mcp"],
"env": {
"ORG_ID": "acme-corp",
"ANTHROPIC_API_KEY": "sk-ant-...",
"SLACK_WEBHOOK_URL": "https://hooks.slack.com/...",
"GITHUB_TOKEN": "ghp_...",
"AUTO_APPLY": "false"
}
}
}
}
Quick start
Running in 15 minutes
Reduce toil. Ship faster. Sleep better.
MIT licensed. Open source. Self-host free or upgrade to Hosted Pro.