Open source · the runtime for AI-built software and agents

Build with AI. Without the decay.

The open-source runtime whose structure keeps AI-generated software coherent in production, instead of collapsing into a mess no one can change. Self-hosted, and yours.

Get started → ★ Star on GitHub
$git clone github.com/wippyai/app
What AI ships works.
Until it falls apart.
Every AI change optimizes for right now, never the whole.

Wippy's answer is constraint. An actor model isolates every part behind messages, so one piece can't tangle another, and what you build holds as it grows, however far you push it.

Structure the AI
can't break.

Left alone, AI code drifts toward entropy. Wippy's structure won't let it: four properties baked into the runtime, not conventions the AI has to follow.

Isolated

No shared state

Parts talk by messages, never by reaching into each other. A change here can't silently break there.

Uniform

One shape for everything

Every part is one kind of registry entry. The AI can't invent five ways to do one thing. Every generation lands in the same mold.

Capability-scoped

Security is a capability

Each part runs with only the capabilities you grant it. No ambient power, nothing it can reach that you didn't hand it.

Self-assembling

No integration tax

New parts register and wire themselves in through their contracts, instead of you gluing services together by hand.

Claude Code and Codex edit it live.

Your agent doesn't guess at Wippy. It reads the real docs and live project state over the Keeper MCP and edits the running system in place.

wippy · build session

See everything it builds.
Roll back anything.

Hand the AI the keys without losing control. Every action it takes is traced and reversible.

Wippy Keeper dashboard: pipeline, dataflows, sessions, and the live registry across all namespaces
keeper · dashboard
Keeper system monitor: the live process tree with hosts, actors, uptime, and state
keeper · system monitor
The dataflow inspector replaying a multi-agent run step by step
keeper · dataflow replay

Not markdown guardrails.
A runtime.

A general-purpose runtime in one self-hosted binary, with the AI toolkit built in. Everything below ships in the box.

wippy · one binary MPL-2.0 · self-hosted
LLMs
Anthropic · OpenAI · Google · local
Agents
memory · delegation
Tools
functions · RAG · MCP
Workflows
durable · Temporal optional
Processes
supervised · crash-isolated
Cluster
multi-node · gossip + Raft
Real-time
WebSocket · SSE · queues
Storage
SQL · vector · files
Policy
capability-scoped
Observability
tracing · OpenTelemetry
Lua + WASM
typed · LSP · linter · tests
Live edit
change it running · roll back

Start from a working app.

Clone the template, point your coding agent at it, and build on top. The structure keeps what it generates coherent.