Feature Requests ● ♾️ WeOwn.Cloud ☁️ ● Version 2 (coming 🔜)

Context: v1 apps (AnythingLLM/n8n) will begin AI calls soon. We must prioritize SSO/KYC and observability first so we can control access and measure usage before traffic ramps and before adding agentic AI features.

1. SSO & KYC / Verifications (Priority 1)

   • SSO (recommend): Keycloak. Alternatives: Unicorn (Keycloak-compatible) or vendor SSO if required. Supports OIDC/SAML to Google/Microsoft; MFA/passkeys optional.
   • KYC/Verifications (recommend): WithPersona.com for identity checks and credential flows.
   • Integration points: WordPress (member login/gating), AnythingLLM (per-tenant keys), n8n (automation), agent stack (RBAC).

2. Observability & Controls (Priority 2)

   • Recommend: Langfuse (traces/evals/prompts) + OpenTelemetry collector + dashboards; enable client SDKs in AnythingLLM, n8n, agents.
   • Budgets/guardrails: per-key quotas, rate limits, routing/fallback at the gateway.
   • Optional PII hygiene: redaction at app or gateway until full DLP is chosen.
   • Alternative: Helicone (self-host) for proxy + analytics in one service with lighter ops.

3. Unified AI Endpoint (Priority 3)

   • Recommend: LiteLLM Proxy as OpenAI-compatible front door for all apps and agents.
   • Alternative: Helicone as an all-in-one gateway if skipping Langfuse/OTel.

4. Inference Topology (Priority 4)

   • Recommend: Hybrid
     • Per-tenant local runtime for light/private calls: Ollama (CPU/tiny GPU) or small vLLM.
     • Route heavy jobs to a central GPU cluster via LiteLLM.
   • Alternative A (maximum isolation): Per-tenant inference in each DOKS cluster using vLLM or TGI. Highest privacy, higher cost.
   • Alternative B (best $/token): One shared GPU cluster with KServe orchestrating vLLM/TGI on GPU nodes; MIG or node affinity for tenancy; budgets at proxy.
   • On-prem option: Boss homelab GPU Kubernetes with NVIDIA GPU Operator, MIG, KServe, vLLM/TGI, LiteLLM; private access via WireGuard/Tailscale.

5. Agent Stack (Priority 5)

   • Core/baseline: kagent.dev for K8s-native ops agents (health checks, backups, cost watchdogs, MCP tools).
   • User/Web3 agents: ElizaOS pilot for wallets/on-chain/public agents.
   • Workflow frameworks (pick per use): LangGraph (DAG/state), CrewAI (role teams), smolagents (lightweight).
   • Advanced template: Claude Agent SDK where we standardize on Anthropic behind our proxy.

6. Self-Hosted Dev & AI Coding Assistants in Kubernetes (Priority 6)

   • Dev runtime sovereignty: Ona as optional self-hosted dev/CDE with MCP (keep DOKS for production).

   • Endpoint-first assistants (default): Serve code models behind LiteLLM; developers use Continue.dev, Aider, or Zed IDE agents.

   • Server-first assistants (in-cluster):

     • Tabby (TabbyML) code completion/chat server.
     • OpenHands self-hosted autonomous coding agent service.
     • kubectl-ai container for DevOps CLI assistance.
     • Sourcegraph Cody Enterprise only if licensing and self-host fit are confirmed.

   • Complements for sovereignty: self-hosted Git (Gitea/Forgejo), container registry (Harbor).

7. Telephony & People Ops Agents (Priority 7)

   • ConnexAI for call center and HR agents (voice/omnichannel).

8. WordPress v2 Performance & DX (Priority 8)

   • Performance: Perfmatters, Redis Object Cache, FastCGI micro-caching, optional CDN.
   • Monitoring: GTmetrix scheduled tests per tenant.
   • DX: Replicable starter theme with landing/opt-in templates wired to FluentCRM tags.

9. GitOps & CI/CD across Clusters and WordPress (Priority 9)

   • Git hosting: Gitea or Forgejo.
   • GitOps: Argo CD (primary). Alternative: FluxCD.
   • CI runners: GitHub Actions self-hosted, GitLab Runner, Drone CI/Woodpecker, Tekton Pipelines.
   • Build in cluster: BuildKit or Kaniko.
   • Release: Helm charts/values for every app (incl. WordPress); promote via Argo CD with environments/approvals.
   • Secrets in GitOps: SOPS+age or Sealed Secrets.
   • WordPress CI/CD: Composer for mu-plugins; theme/plugin repos; image builds; Argo CD app-of-apps for multi-tenant rollout.

   How to Choose

   • Shipping AI calls now → do SSO/KYC and Observability first, then the gateway.
   • Best cost per token at scale → shared GPU cluster (KServe + vLLM/TGI) routed via LiteLLM.
   • Strict isolation/compliance → per-tenant inference or MIG-partitioned slices with hard budgets.
   • Small tenants/cohorts → Hybrid: local Ollama for light tasks + shared GPU for heavy.
   • Public/Web3 agents → ElizaOS; internal ops → kagent.dev; complex workflows → LangGraph/CrewAI; tiny automations → smolagents.
   • Coding help with minimal ops → endpoint-first; team UX/central control → Tabby.
   • Sovereign dev envs → Ona add-on; keep DOKS for production.