# Katakate

- [Katakate](https://docs.katakate.org/index.md)
- [k7](https://docs.katakate.org/k7/index.md): Secure sandboxed compute for AI agents and workloads
- [Quickstart](https://docs.katakate.org/k7/getting-started/installation.md): Install k7, prepare your node, start the API, and run your first sandbox
- [Architecture](https://docs.katakate.org/k7/concepts/architecture.md): How the CLI, API, K3s, Kata, k7d, and the four VM backends fit together
- [Backends](https://docs.katakate.org/k7/concepts/backends.md): kfd vs kql vs k7d vs k7d-fc — how to choose
- [Networking](https://docs.katakate.org/k7/concepts/networking.md): Cilium, platform isolation, opt-in ingress and NodePort exposure, FQDN-aware egress whitelists, Hubble
- [Security model](https://docs.katakate.org/k7/concepts/security-model.md): VM isolation, jailer, capabilities, non-root, and API key handling
- [CLI reference](https://docs.katakate.org/k7/guides/cli.md): All k7 commands with options and examples
- [Python SDK](https://docs.katakate.org/k7/guides/python-sdk.md): Use the k7-sdk package to manage sandboxes from Python
- [Multi-node clusters](https://docs.katakate.org/k7/guides/multi-node.md): Inventory format, HA, dual-backend nodes, and Longhorn replication
- [Snapshots & fork](https://docs.katakate.org/k7/guides/snapshots-fork.md): Pause, resume, and fork sandboxes via crash-consistent VolumeSnapshots
- [Docker in a sandbox](https://docs.katakate.org/k7/guides/docker.md): k7 create --docker — a pinned dockerd with overlay2 on a block disk, on every backend
- [LangChain agent tutorial](https://docs.katakate.org/k7/guides/langchain-agent.md): Build a ReAct agent that executes inside a K7 sandbox
- [Utilities](https://docs.katakate.org/k7/guides/utilities.md): Helper scripts: disk wipe for thin‑pool prep and high‑density stress testing
- [Auth & responses](https://docs.katakate.org/k7/api/introduction.md): Authentication, base URL, and response format
- [Security & networking](https://docs.katakate.org/k7/api/security.md): Non-root execution, Linux capabilities, and egress control
- [Sandboxes](https://docs.katakate.org/k7/api/endpoints/sandboxes.md): Create, list, get, and delete sandboxes
- [Snapshots](https://docs.katakate.org/k7/api/endpoints/snapshots.md): Inspect, create, delete and GC VolumeSnapshots
- [Execute command](https://docs.katakate.org/k7/api/endpoints/exec.md): Run a command inside a sandbox
- [Metrics](https://docs.katakate.org/k7/api/endpoints/metrics.md): Get CPU and memory usage for sandboxes
- [Nodes](https://docs.katakate.org/k7/api/endpoints/nodes.md): Per-node storage-pool utilization
- [Health](https://docs.katakate.org/k7/api/endpoints/health.md): Root and /health endpoints
- [K7d](https://docs.katakate.org/k7d/index.md): Fork a live Kubernetes cluster. Two clocks: ~105 ms (minimal CI guest) and ~1.1 s (Ubuntu k3s+Cilium+Tetragon). Then warp the guest clock so waiting is free.
- [Installation](https://docs.katakate.org/k7d/getting-started/installation.md): Five-minute quickstart, install from a release tarball or build from source, then fork your first cluster
- [CoW fork](https://docs.katakate.org/k7d/concepts/cow-fork.md): How a ~5 ms fork of a running VM is possible: MAP_PRIVATE, dirty-page tracking, and what had to be right
- [Network identity](https://docs.katakate.org/k7d/concepts/network-identity.md): Why a forked cluster doesn't reboot: per-fork bridges with replayed IPs and MACs, jailed from each other
- [Snapshot tree](https://docs.katakate.org/k7d/concepts/snapshot-tree.md): Forks live in a daemon-managed tree with RAM/disk budgets, LRU eviction, and O(divergence) reflink keyframes
- [Architecture](https://docs.katakate.org/k7d/concepts/architecture.md): The k7d stack: daemon, containerd shim, guest agent, two guests, two VMM engines — who owns what
- [Backends: native and Firecracker](https://docs.katakate.org/k7d/concepts/backends.md): One daemon, one shim, two VMMs — k7d's in-process rust-vmm engine or stock Firecracker under the stock jailer (RuntimeClass k7-fc)
- [Security model](https://docs.katakate.org/k7d/concepts/security.md): What k7d isolates, what it deliberately doesn't, what the tree jailer contains, the Firecracker profile, and current limitations
- [GRPO / agent tree search](https://docs.katakate.org/k7d/guides/grpo.md): Wire k7d into a GRPO trainer, a tree-search agent, or an MCP-driven frontier model: fork_batch, protect, prune, auto_evict
- [Cluster mode](https://docs.katakate.org/k7d/guides/cluster-mode.md): Create, adopt, and fork whole multi-VM Kubernetes clusters as one unit
- [Time warp](https://docs.katakate.org/k7d/guides/time-warp.md): Only waiting is compressible: lockstep guest-clock jumps, auto-warp, and continuous ×N KVM time dilation
- [Ubuntu nodes](https://docs.katakate.org/k7d/guides/ubuntu-nodes.md): Stock Ubuntu 24.04 guests as bare k7d VMs: build the node image, boot a k3s cluster with Cilium, Tetragon, or Longhorn, and fork it live
- [RuntimeClass k7 and k7-fc](https://docs.katakate.org/k7d/guides/runtime-class.md): Run Kubernetes pods as k7d microVMs: daemon, shim, containerd config, RuntimeClass, and the Firecracker-backed k7-fc class
- [Feature matrix](https://docs.katakate.org/k7d/guides/feature-matrix.md): Kubernetes feature support — inside a forked cluster, and for pods running as k7d VMs
- [Daemon API](https://docs.katakate.org/k7d/api/protocol.md): JSON-lines protocol over /run/k7d/k7d.sock — VM, tree, clock, and host operations
- [Benchmarks](https://docs.katakate.org/k7d/deep-dives/benchmarks.md): Named hardware, exact commands, and enforced latency budgets for every claim — both fork clocks, keyframes, and time warp
- [Formal verification](https://docs.katakate.org/k7d/deep-dives/formal-verification.md): Kani bounded model checking and Aeneas→Lean proofs on the paths where they pay off


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.