Roadmap & FAQ
Full reference: README, what's in v1.0 and what's next in the repository.
Roadmap
| Area | Ready in v1.0 | Next (v1.1+) |
|---|---|---|
| Clusters | Hub and agents over outbound WebSocket; Add cluster wizard with install guide, one-time join tokens and a live connection checklist; 1..N agent replicas per cluster | mTLS for agents |
| Flux | Kustomization, HelmRelease, Git/OCI/Helm repositories, HelmChart, Bucket; Karpenter and External Secrets presets; workloads, pods and core kinds; inventory with YAML and events | Image automation, notification alerts, flux diff view |
| Actions | Reconcile (with source), suspend, resume; typed confirmation on protected clusters | Workload restart, bulk actions |
| UI | Keyboard-first SPA, ⌘K palette with fleet-wide search, live dependency graph, detail views (YAML, events, workload logs), environment colours, dark mode | Saved views |
| Sign-in | GitHub (OAuth App or GitHub App) and OIDC (Google, Okta, Entra ID, Dex); local users, optionally break-glass only; trusted reverse-proxy headers | Hub-managed access (assign users to groups, expiring grants). Future: SAML 2.0 |
| Access tokens | Personal access tokens (read / operate, mandatory expiry) for MCP | MCP OAuth 2.1, per-cluster scoping |
| MCP | /mcp with fleet-wide read tools, guarded actions and thread tools | Log follow, subscriptions |
| Threads | Review threads on any resource or cluster, from people, Claude Code and Ask AI | Mentions, notifications, webhooks |
| Ask AI | Anthropic API, or AWS Bedrock with Converse models that support tool use (Claude tested; IRSA or Pod Identity, Guardrails); read-only tools, redaction, log and YAML attachments | OpenAI-compatible endpoints (Ollama, vLLM, Azure OpenAI), streaming answers |
| Storage and HA | PostgreSQL (bring your own, for example CloudNativePG); multiple hub replicas, active/active | Read replicas |
| Ops | Helm charts, internal ingress examples, audit log, runtime kill switches | Prometheus dashboards, audit webhook |
The full table is in the README. Plans can change.
FAQ
What is the status?
Eddy is v1.0, an early release. It already includes GitHub and OIDC sign-in and active/active hub replicas on PostgreSQL, which you provide. mTLS for agents, MCP OAuth and hub-managed access are planned next.
Do you need to expose workload clusters?
No. Agents dial out to the hub. The hub holds no workload-cluster credentials.
Can Eddy read my Secrets?
No. The agent does not watch Secrets or ConfigMaps, and their data never leaves a cluster.
Who can do what?
Your Kubernetes RBAC decides. Eddy runs each request as the signed-in user through impersonation, and invents no roles of its own.
Is AI required?
No. Ask AI is off by default, read-only, and works with the Anthropic API or AWS Bedrock. MCP is on by default; turn it off with config.mcp.enabled: false. Both can also be switched off at runtime.
Can I use my company login?
Yes. Eddy signs in with GitHub or OIDC (Google, Okta, Entra, Dex) natively, and a trusted reverse proxy such as oauth2-proxy also works. SAML 2.0 is a possible future addition. See Sign-in.
How do I contribute or report a problem?
Read CONTRIBUTING.md and open an issue on GitHub. For vulnerabilities, see SECURITY.md.