githubinferredactive
ai-homebase
provenance:github:sbischoff-ai/ai-homebase
A ready-to-deploy Kubernetes-based setup for an all-purpose AI assistant, coding agent and complete homelab based on OpenClaw.
README
# ai-homebase `ai-homebase` is an opinionated homelab platform for running OpenClaw as the center of a personal AI control plane. It combines a multi-agent OpenClaw setup with supporting services such as Nextcloud, a dedicated Nextcloud MCP gateway, Gitea, an in-cluster Docker registry, Paperless-ngx, Vaultwarden, shared PostgreSQL/Redis, an in-cluster Postfix relay for application email, shared Qdrant/Qdrant MCP memory services for cross-agent RAG context, and Memgraph plus Memgraph Lab for graph-structured long-term knowledge. The point of the repo is not just “a pile of charts.” It gives you one coherent platform shape that works in two places: `k3d` for fast local iteration and `k3s` for the long-running homelab deployment. The bootstrap flow, secrets model, hostnames, multi-agent topology, MCP posture, and service contracts stay aligned between those targets so local validation is actually useful before you touch the real server. Characteristic features of this stack: - OpenClaw bootstrapped as a multi-agent system with `main`, `architect`, `coder`, `archivist`, and `watchdog` - Architect-oriented project documentation seeded into Nextcloud so the cluster can document and evolve itself from day one - remote Docker sandboxes for specialist agents, including a coder-specific sandbox image with developer tooling - Nextcloud shared with OpenClaw through a dedicated MCP gateway and shared-account operating conventions - GitOps handoff into an in-cluster Gitea repository with Argo CD bootstrapped, initially synced, and then kept in manual-sync mode - an in-cluster authenticated registry for coder-built application images - one bootstrap input model for both local and homelab targets through `bootstrap.local.toml` The intended long-running target today is a single-node `k3s` install on a Hetzner A42U-class machine with a Ryzen 7 Pro 8700GE, 64 GB RAM, and roughly 3 TB of storage. The repo now assumes that target should have headroom not only for the current stack, but also for additional heavier services such as Qdrant, Memgraph, and future coder-deployed web services. ## Start Here - Target chooser: [docs/deployment.md](./docs/deployment.md) - Full docs index: [docs/README.md](./docs/README.md) - Configuration and bootstrap model: [docs/configuration.md](./docs/configuration.md) - Service contracts: [docs/services.md](./docs/services.md) - Secret workflow: [docs/secrets.md](./docs/secrets.md) ## Quick Start ### Local `k3d` ```bash cp bootstrap.example.toml bootstrap.local.toml ./scripts/k3d-local-bootstrap.sh --cluster-name ai-homebase-dev --bootstrap-config bootstrap.local.toml ``` Use [docs/deployment-k3d.md](./docs/deployment-k3d.md) for the full local workflow, including the required NixOS host preparation and the integrated GitOps handoff that now completes before the bootstrap returns. ### Homelab `k3s` ```bash sudo ./scripts/install-k3s-ubuntu-2404.sh cp bootstrap.example.toml bootstrap.local.toml ./scripts/bootstrap-stack.sh --profile k3s --bootstrap-config bootstrap.local.toml ``` Use [docs/runbook-homelab.md](./docs/runbook-homelab.md) for the full Ubuntu host-prep and integrated bootstrap path. In both cases, fill in the new `[mail]` section and the per-agent model sections in `bootstrap.local.toml` before bootstrapping so Nextcloud/Vaultwarden mail and the bootstrapped OpenClaw `main`, `architect`, `coder`, `archivist`, and `watchdog` agents are configured correctly. Each agent supports `model` plus optional `fallback_models` in the bootstrap config. Provider tokens and the migrated application Secrets should now be managed through the SOPS workflow documented in [docs/secrets.md](./docs/secrets.md) rather than committed in values files or created imperatively during bootstrap. The default coder posture now assumes a Claude-based orchestrator delegating substantial coding to Codex, so the standard bootstrap expects both `ANTHROPIC_API_KEY` and `OPENAI_API_KEY` to be present through the `openclaw-secrets` Secret. The same bootstrap flow also creates a dedicated `openclaw` Nextcloud user, seeds the OpenClaw gateway config with the standard Nextcloud MCP server definition, pre-seeds specialist workspace files for the multi-agent topology, and seeds the initial Memgraph knowledge graph baseline for the cluster. ## Documentation Map - Target guides: [docs/deployment-k3d.md](./docs/deployment-k3d.md), [docs/runbook-homelab.md](./docs/runbook-homelab.md) - Deep dives: [docs/architecture.md](./docs/architecture.md), [docs/security.md](./docs/security.md), [docs/networking.md](./docs/networking.md), [docs/gitops.md](./docs/gitops.md) - Operational reference: [docs/commands.md](./docs/commands.md), [docs/services.md](./docs/services.md), [docs/storage.md](./docs/storage.md) - Troubleshooting: [docs/k3d-troubleshooting.md](./docs/k3d-troubleshooting.md)
PUBLIC HISTORY
First discoveredApr 2, 2026
TRUST SCORE
32/ 100
• No PROVENANCE.yml0
• Classifier confidence+15
• Age (20d old)+2
✓ No incidents+15
• No crypto identity0
METADATA
platformgithub
first seenMar 12, 2026
last updatedApr 1, 2026
last crawledtoday
version—
confidence50%
CONFIDENCE
README BADGE
Add to your README:
